Image Generation from Layout

Bo Zhao Lili Meng Weidong Yin Leonid Sigal

University of British Columbia Vector Institute

{bzhao03, menglili, wdyin, lsigal}@cs.ubc.ca

Abstract

Despite significant recent progress on generative mod-

els, controlled generation of images depicting multiple and

complex object layouts is still a difficult problem. Among

the core challenges are the diversity of appearance a given

object may possess and, as a result, exponential set of im-

ages consistent with a specified layout. To address these

challenges, we propose a novel approach for layout-based

image generation; we call it Layout2Im. Given the coarse

spatial layout (bounding boxes + object categories), our

model can generate a set of realistic images which have

the correct objects in the desired locations. The represen-

tation of each object is disentangled into a specified/certain

part (category) and an unspecified/uncertain part (appear-

ance). The category is encoded using a word embedding

and the appearance is distilled into a low-dimensional vec-

tor sampled from a normal distribution. Individual object

representations are composed together using convolutional

LSTM, to obtain an encoding of the complete layout, and

then decoded to an image. Several loss terms are intro-

duced to encourage accurate and diverse image generation.

The proposed Layout2Im model significantly outperforms

the previous state-of-the-art, boosting the best reported in-

ception score by 24.66% and 28.57% on the very challeng-

ing COCO-Stuff and Visual Genome datasets, respectively.

Extensive experiments also demonstrate our model’s ability

to generate complex and diverse images with many objects.

1. Introduction

Image generation of complex realistic scenes with mul-

tiple objects and desired layouts is one of the core fron-

tiers for computer vision. Existence of such algorithms

would not only inform our designs for inference mecha-

nisms, needed for visual understanding, but also provide

practical application benefits in terms of automatic image

generation for artists and users. In fact, such algorithms, if

successful, may replace visual search and retrieval engines

in their entirety. Why search the web for an image, if you

can create one to user specification?

Image

Decoder

Object

Encoder

Objects

Fuserman

boy

frisbee man

Object

Composer

N (zs)<latexit sha1_base64="OQOr6Z79+QTFi3Zp2TyJ/nBKqwc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIui25cSQV7gTaUyXTSDp1MwsxEqCG48VXcuFDErU/hzrdxkmahrT8MfPznHOac34sYlcq2v43S0vLK6lp5vbKxubW9Y+7utWUYC0xaOGSh6HpIEkY5aSmqGOlGgqDAY6TjTa6yeueeCElDfqemEXEDNOLUpxgpbQ3Mg36A1BgjltyktZw9P3lIB/JkYFbtup3LWgSngCoUag7Mr/4wxHFAuMIMSdlz7Ei5CRKKYkbSSj+WJEJ4gkakp5GjgEg3yU9IrWPtDC0/FPpxZeXu74kEBVJOA093ZkvK+Vpm/lfrxcq/cBPKo1gRjmcf+TGzVGhleVhDKghWbKoBYUH1rhYeI4Gw0qlVdAjO/MmL0D6tO5pvz6qNyyKOMhzCEdTAgXNowDU0oQUYHuEZXuHNeDJejHfjY9ZaMoqZffgj4/MHjumXhA==</latexit><latexit sha1_base64="OQOr6Z79+QTFi3Zp2TyJ/nBKqwc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIui25cSQV7gTaUyXTSDp1MwsxEqCG48VXcuFDErU/hzrdxkmahrT8MfPznHOac34sYlcq2v43S0vLK6lp5vbKxubW9Y+7utWUYC0xaOGSh6HpIEkY5aSmqGOlGgqDAY6TjTa6yeueeCElDfqemEXEDNOLUpxgpbQ3Mg36A1BgjltyktZw9P3lIB/JkYFbtup3LWgSngCoUag7Mr/4wxHFAuMIMSdlz7Ei5CRKKYkbSSj+WJEJ4gkakp5GjgEg3yU9IrWPtDC0/FPpxZeXu74kEBVJOA093ZkvK+Vpm/lfrxcq/cBPKo1gRjmcf+TGzVGhleVhDKghWbKoBYUH1rhYeI4Gw0qlVdAjO/MmL0D6tO5pvz6qNyyKOMhzCEdTAgXNowDU0oQUYHuEZXuHNeDJejHfjY9ZaMoqZffgj4/MHjumXhA==</latexit><latexit sha1_base64="OQOr6Z79+QTFi3Zp2TyJ/nBKqwc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIui25cSQV7gTaUyXTSDp1MwsxEqCG48VXcuFDErU/hzrdxkmahrT8MfPznHOac34sYlcq2v43S0vLK6lp5vbKxubW9Y+7utWUYC0xaOGSh6HpIEkY5aSmqGOlGgqDAY6TjTa6yeueeCElDfqemEXEDNOLUpxgpbQ3Mg36A1BgjltyktZw9P3lIB/JkYFbtup3LWgSngCoUag7Mr/4wxHFAuMIMSdlz7Ei5CRKKYkbSSj+WJEJ4gkakp5GjgEg3yU9IrWPtDC0/FPpxZeXu74kEBVJOA093ZkvK+Vpm/lfrxcq/cBPKo1gRjmcf+TGzVGhleVhDKghWbKoBYUH1rhYeI4Gw0qlVdAjO/MmL0D6tO5pvz6qNyyKOMhzCEdTAgXNowDU0oQUYHuEZXuHNeDJejHfjY9ZaMoqZffgj4/MHjumXhA==</latexit><latexit sha1_base64="OQOr6Z79+QTFi3Zp2TyJ/nBKqwc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIui25cSQV7gTaUyXTSDp1MwsxEqCG48VXcuFDErU/hzrdxkmahrT8MfPznHOac34sYlcq2v43S0vLK6lp5vbKxubW9Y+7utWUYC0xaOGSh6HpIEkY5aSmqGOlGgqDAY6TjTa6yeueeCElDfqemEXEDNOLUpxgpbQ3Mg36A1BgjltyktZw9P3lIB/JkYFbtup3LWgSngCoUag7Mr/4wxHFAuMIMSdlz7Ei5CRKKYkbSSj+WJEJ4gkakp5GjgEg3yU9IrWPtDC0/FPpxZeXu74kEBVJOA093ZkvK+Vpm/lfrxcq/cBPKo1gRjmcf+TGzVGhleVhDKghWbKoBYUH1rhYeI4Gw0qlVdAjO/MmL0D6tO5pvz6qNyyKOMhzCEdTAgXNowDU0oQUYHuEZXuHNeDJejHfjY9ZaMoqZffgj4/MHjumXhA==</latexit>

Figure 1. Image generation from layout. Given the coarse lay-

out (bounding boxes + object categories), the proposed Layout2Im

model samples the appearance of each object from a normal distri-

bution, and transforms these inputs into a real image by a serial of

components. Please refer to Section 3 for a detailed explanation.

For these reasons, image generation algorithms have

been a major focus of recent research. Of specific relevance

are approaches for text-to-image [11, 15, 25, 33, 41, 47]

generation. By allowing users to describe visual concepts

in natural language, text-to-image generation provides nat-

ural and flexible interface for conditioned image genera-

tion. However, existing text-to-image approaches exhibit

two drawbacks: (i) most approaches can only generate plau-

sible results on simple datasets such as cats [49], birds [44]

or flowers [30]. Generating complex, real-world images

such as those in COCO-Stuff [1] and Visual Genome [19]

datasets remains a challenge; (ii) the ambiguity of textual

description makes it more difficult to constrain complex

generation process, e.g., locations and sizes of different ob-

jects are usually not given in the description.

Scene graphs are powerful structured representations

that encode objects, their attributes and relationships.

In [14] an approach for generating complex images with

many objects and relationships is proposed by conditioning

the generation on scene graphs. It addresses some of the

aforementioned challenges. However, scene graphs are dif-

ficult to construct for a layman user and lack specification

of core spatial properties, e.g., object size / position.

To overcome these limitations, we propose to generate

complicated real-world images from layouts, as illustrated

in Figure 1. By simply specifying the coarse layout (bound-

ing boxes + categories) of the expected image, our proposed

model can generate an image which contains the desired ob-

18584

jects in the correct locations. It is much more controllable

and flexible to generate an image from layout than textual

description.

With the new task comes new challenges. First, image

generation from layout is a difficult one-to-many problem.

Many images could be consistent with a specified layout;

same layout may be realized by different appearance of

objects, or even their interactions (e.g., a person next to

the frisbee may be throwing it or be a bystander, see Fig-

ure 1). Second, the information conveyed by a bounding

box and corresponding label is very limited. The actual ap-

pearance of the object displayed in an image is not only

determined by its category and location, but also its inter-

actions and consistency with other objects. Moreover, spa-

tially close objects may have overlapping bounding boxes.

This leads to additional challenges of “separating” which

object should contribute to individual pixels. A good gener-

ative model should take all these factors and challenges into

account implicitly or explicitly.

We address these challenges using a novel variational in-

ference approach. The representation of each object in the

image is explicitly disentangled into a specified/certain part

(category) and an unspecified/uncertain part (appearance).

The category is encoded using a word embedding and the

appearance is distilled into a low-dimensional vector sam-

pled from a normal distribution. Based on this representa-

tion and specification of object bounding box, we construct

a feature map for each object. These feature maps are then

composed using convolutional LSTM into a hidden feature

map for the entire image, which subsequently is decoded

into an output image. This set of modelling choices makes

it easy to generate different and diverse images by sampling

the appearance of individual objects, and/or adding, moving

or deleting objects from the layout. Our proposed model is

end-to-end learned using a loss that consists of a number

of objectives. Specifically, a pair of discriminators are de-

signed to discriminate the overall generated image and the

generated objects within their specified bounding boxes, as

real or fake. In addition, object discriminator is also trained

to classify the categories of generated objects.

Contributions. Our contributions are three-fold: (1) We

propose a novel approach for generating images from coarse

layout (bounding boxes + object categories). This provides

a flexible control mechanism for image generation. (2) By

disentangling the representation of objects into a category

and (sampled) appearance, our model is capable of generat-

ing a diverse set of consistent images from the same layout.

(3) We show qualitative and quantitative results on COCO-

Stuff [1] and Visual Genome [19] datasets, demonstrating

our model’s ability to generate complex images with respect

to object categories and their layout (without access to seg-

mentation masks [11, 14]). We also preform comprehensive

ablations to validate each component in our approach.

2. Related Work

Conditional Image Generation. Conditional image gen-

eration approaches generate images conditioned on addi-

tional input information, including entire source image [13,

23, 32, 46, 50, 51, 52], sketches [13, 36, 43, 45, 52], scene

graphs [14], dialogues [16, 37] and text descriptions [25, 33,

41, 47]. Variational Autoencoders (VAEs) [18, 25, 40], au-

toregressive models [31, 42] and GANs [13, 27, 43, 51] are

powerful tools for conditional image generation and have

shown promising results. However, many previouse gener-

ative models [13, 32, 36, 45, 46, 51] tend to largely ignore

the random noise vector when conditioning on the same rel-

evant context, making the generated images very similar to

each other. By enforcing the bijection mapping between the

latent and target space, BicycleGAN [52] pursues the diver-

sity of generated images from a same input. Inspired by this

idea, in our paper, we also explicitly regress the latent codes

which are used to generate the different objects.

Image Generation from Layout. The use of layout in

image generation is a relative novel task. In prior works, it

is usually served as an intermediate representation between

other input sources (e.g., text [11] or scene graphs [14]) and

the output images, or as a complementary feature for image

generation based on context (e.g., text [15, 34, 41], shape

and lighting [6]). In [11, 14], instead of learning a direct

mapping from textual description/scene graph to an image,

the generation process is decomposed into multiple individ-

ual steps. They first construct a semantic layout (bound-

ing boxes + object shapes) from the input, and then con-

vert it to an image using an image generator. Both of them

can generate an image from a coarse layout together with

textual description/scene graph. However [11] requires de-

tailed object instance segmentation masks to train its ob-

ject shape generator. Getting such segmentation masks for

large scale datasets is both time-consuming and and labor-

intensive. Different from [11] and [14], we use the coarse

layout without instance segmentation mask as a fundamen-

tal input modality for diverse image generation.

Disentangled Representations. Many papers [2, 3, 5, 21,

22, 24, 26, 29] have tried to learn disentangled representa-

tions as part of image generation. Disentangled represen-

tations model different factors of data variations, such as

class-related and class-independent parts [3, 21, 22, 26, 29].

By manipulating the disentangled representations, images

with different appearances can be generated easily. In [24],

three factors (foreground, background and pose) are dis-

entangled explicitly when generating person image. In-

foGAN [2], DrNet [5] and DRIT [22] learn the disen-

tangled representations in an unsupervised manner, either

by maximizing the mutual information [2] or adversarial

losses [5, 22]. In our work, we explicitly separate the rep-

resentation of each object into a category-related and an

8585

ObjectEstimator

Image

Decoder

Object

Encoder

Objects

Fuser

D img

D obj

D img

L1

L1

KL

Image Reconstruction Path

Random Image Generation Path

Multiple Inputs

Recurrent Cells

Sample from Distirbution

Loss

Deep Networks

Loss Path ObjectEstimator

Object

Composer

I0

<latexit sha1_base64="Sbo2ww/GDIm9l8j5YjXsFQN4ZvE=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbc1F0F+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdj+7GFSqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMbzpEk</latexit><latexit sha1_base64="Sbo2ww/GDIm9l8j5YjXsFQN4ZvE=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbc1F0F+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdj+7GFSqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMbzpEk</latexit><latexit sha1_base64="Sbo2ww/GDIm9l8j5YjXsFQN4ZvE=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbc1F0F+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdj+7GFSqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMbzpEk</latexit><latexit sha1_base64="Sbo2ww/GDIm9l8j5YjXsFQN4ZvE=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbc1F0F+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdj+7GFSqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMbzpEk</latexit>

I<latexit sha1_base64="Z8trNEa2Pi7+nrAVHObru6CLqps=">AAAB+XicbVBNS8NAFHypX7V+RT16CRbBU0lE0GPBS71VsLXQlLLZbtqlm03YfSmUkH/ixYMiXv0n3vw3btoctHVgYZh5jzc7QSK4Rtf9tiobm1vbO9Xd2t7+weGRfXzS1XGqKOvQWMSqFxDNBJesgxwF6yWKkSgQ7CmY3hX+04wpzWP5iPOEDSIyljzklKCRhrbtRwQnQZj5E4LZfZ4P7brbcBdw1olXkjqUaA/tL38U0zRiEqkgWvc9N8FBRhRyKlhe81PNEkKnZMz6hkoSMT3IFslz58IoIyeMlXkSnYX6eyMjkdbzKDCTRU696hXif14/xfB2kHGZpMgkXR4KU+Fg7BQ1OCOuGEUxN4RQxU1Wh06IIhRNWTVTgrf65XXSvWp4hj9c15utso4qnME5XIIHN9CEFrShAxRm8Ayv8GZl1ov1bn0sRytWuXMKf2B9/gAQCJPx</latexit><latexit sha1_base64="Z8trNEa2Pi7+nrAVHObru6CLqps=">AAAB+XicbVBNS8NAFHypX7V+RT16CRbBU0lE0GPBS71VsLXQlLLZbtqlm03YfSmUkH/ixYMiXv0n3vw3btoctHVgYZh5jzc7QSK4Rtf9tiobm1vbO9Xd2t7+weGRfXzS1XGqKOvQWMSqFxDNBJesgxwF6yWKkSgQ7CmY3hX+04wpzWP5iPOEDSIyljzklKCRhrbtRwQnQZj5E4LZfZ4P7brbcBdw1olXkjqUaA/tL38U0zRiEqkgWvc9N8FBRhRyKlhe81PNEkKnZMz6hkoSMT3IFslz58IoIyeMlXkSnYX6eyMjkdbzKDCTRU696hXif14/xfB2kHGZpMgkXR4KU+Fg7BQ1OCOuGEUxN4RQxU1Wh06IIhRNWTVTgrf65XXSvWp4hj9c15utso4qnME5XIIHN9CEFrShAxRm8Ayv8GZl1ov1bn0sRytWuXMKf2B9/gAQCJPx</latexit><latexit sha1_base64="Z8trNEa2Pi7+nrAVHObru6CLqps=">AAAB+XicbVBNS8NAFHypX7V+RT16CRbBU0lE0GPBS71VsLXQlLLZbtqlm03YfSmUkH/ixYMiXv0n3vw3btoctHVgYZh5jzc7QSK4Rtf9tiobm1vbO9Xd2t7+weGRfXzS1XGqKOvQWMSqFxDNBJesgxwF6yWKkSgQ7CmY3hX+04wpzWP5iPOEDSIyljzklKCRhrbtRwQnQZj5E4LZfZ4P7brbcBdw1olXkjqUaA/tL38U0zRiEqkgWvc9N8FBRhRyKlhe81PNEkKnZMz6hkoSMT3IFslz58IoIyeMlXkSnYX6eyMjkdbzKDCTRU696hXif14/xfB2kHGZpMgkXR4KU+Fg7BQ1OCOuGEUxN4RQxU1Wh06IIhRNWTVTgrf65XXSvWp4hj9c15utso4qnME5XIIHN9CEFrShAxRm8Ayv8GZl1ov1bn0sRytWuXMKf2B9/gAQCJPx</latexit><latexit sha1_base64="Z8trNEa2Pi7+nrAVHObru6CLqps=">AAAB+XicbVBNS8NAFHypX7V+RT16CRbBU0lE0GPBS71VsLXQlLLZbtqlm03YfSmUkH/ixYMiXv0n3vw3btoctHVgYZh5jzc7QSK4Rtf9tiobm1vbO9Xd2t7+weGRfXzS1XGqKOvQWMSqFxDNBJesgxwF6yWKkSgQ7CmY3hX+04wpzWP5iPOEDSIyljzklKCRhrbtRwQnQZj5E4LZfZ4P7brbcBdw1olXkjqUaA/tL38U0zRiEqkgWvc9N8FBRhRyKlhe81PNEkKnZMz6hkoSMT3IFslz58IoIyeMlXkSnYX6eyMjkdbzKDCTRU696hXif14/xfB2kHGZpMgkXR4KU+Fg7BQ1OCOuGEUxN4RQxU1Wh06IIhRNWTVTgrf65XXSvWp4hj9c15utso4qnME5XIIHN9CEFrShAxRm8Ayv8GZl1ov1bn0sRytWuXMKf2B9/gAQCJPx</latexit>

O<latexit sha1_base64="DAwFCmBrqdRCclPO3g01PNx1rZ0=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRbcdGcF+8A2lMl00g6dTMLMjVBC/8KNC0Xc+jfu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmt7nfeeLaiFg94DThfkRHSoSCUbTSYz+iOA7C7G42qFTdmjsHWSVeQapQoDmofPWHMUsjrpBJakzPcxP0M6pRMMln5X5qeELZhI54z1JFI278bJ54Rs6tMiRhrO1TSObq742MRsZMo8BO5gnNspeL/3m9FMMbPxMqSZErtvgoTCXBmOTnk6HQnKGcWkKZFjYrYWOqKUNbUtmW4C2fvEralzXP8vurar1R1FGCUziDC/DgGurQgCa0gIGCZ3iFN8c4L86787EYXXOKnRP4A+fzB8C8kPk=</latexit><latexit sha1_base64="DAwFCmBrqdRCclPO3g01PNx1rZ0=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRbcdGcF+8A2lMl00g6dTMLMjVBC/8KNC0Xc+jfu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmt7nfeeLaiFg94DThfkRHSoSCUbTSYz+iOA7C7G42qFTdmjsHWSVeQapQoDmofPWHMUsjrpBJakzPcxP0M6pRMMln5X5qeELZhI54z1JFI278bJ54Rs6tMiRhrO1TSObq742MRsZMo8BO5gnNspeL/3m9FMMbPxMqSZErtvgoTCXBmOTnk6HQnKGcWkKZFjYrYWOqKUNbUtmW4C2fvEralzXP8vurar1R1FGCUziDC/DgGurQgCa0gIGCZ3iFN8c4L86787EYXXOKnRP4A+fzB8C8kPk=</latexit><latexit sha1_base64="DAwFCmBrqdRCclPO3g01PNx1rZ0=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRbcdGcF+8A2lMl00g6dTMLMjVBC/8KNC0Xc+jfu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmt7nfeeLaiFg94DThfkRHSoSCUbTSYz+iOA7C7G42qFTdmjsHWSVeQapQoDmofPWHMUsjrpBJakzPcxP0M6pRMMln5X5qeELZhI54z1JFI278bJ54Rs6tMiRhrO1TSObq742MRsZMo8BO5gnNspeL/3m9FMMbPxMqSZErtvgoTCXBmOTnk6HQnKGcWkKZFjYrYWOqKUNbUtmW4C2fvEralzXP8vurar1R1FGCUziDC/DgGurQgCa0gIGCZ3iFN8c4L86787EYXXOKnRP4A+fzB8C8kPk=</latexit><latexit sha1_base64="DAwFCmBrqdRCclPO3g01PNx1rZ0=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRbcdGcF+8A2lMl00g6dTMLMjVBC/8KNC0Xc+jfu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmt7nfeeLaiFg94DThfkRHSoSCUbTSYz+iOA7C7G42qFTdmjsHWSVeQapQoDmofPWHMUsjrpBJakzPcxP0M6pRMMln5X5qeELZhI54z1JFI278bJ54Rs6tMiRhrO1TSObq742MRsZMo8BO5gnNspeL/3m9FMMbPxMqSZErtvgoTCXBmOTnk6HQnKGcWkKZFjYrYWOqKUNbUtmW4C2fvEralzXP8vurar1R1FGCUziDC/DgGurQgCa0gIGCZ3iFN8c4L86787EYXXOKnRP4A+fzB8C8kPk=</latexit>

H<latexit sha1_base64="y6UIlSXRZgD9e27DiUCwLf/y8P0=">AAAB8XicbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9oFtKJPppB06mYSZG6GE/oUbF4q49W/c+TdO2iy09cDA4Zx7mXNPkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeLaiFg94CzhfkTHSoSCUbTS4yCiOAnCrDkfVmtu3V2ArBOvIDUo0BpWvwajmKURV8gkNabvuQn6GdUomOTzyiA1PKFsSse8b6miETd+tkg8JxdWGZEw1vYpJAv190ZGI2NmUWAn84Rm1cvF/7x+iuGtnwmVpMgVW34UppJgTPLzyUhozlDOLKFMC5uVsAnVlKEtqWJL8FZPXiedq7pn+f11rdEs6ijDGZzDJXhwAw1oQgvawEDBM7zCm2OcF+fd+ViOlpxi5xT+wPn8AbYZkPI=</latexit><latexit sha1_base64="y6UIlSXRZgD9e27DiUCwLf/y8P0=">AAAB8XicbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9oFtKJPppB06mYSZG6GE/oUbF4q49W/c+TdO2iy09cDA4Zx7mXNPkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeLaiFg94CzhfkTHSoSCUbTS4yCiOAnCrDkfVmtu3V2ArBOvIDUo0BpWvwajmKURV8gkNabvuQn6GdUomOTzyiA1PKFsSse8b6miETd+tkg8JxdWGZEw1vYpJAv190ZGI2NmUWAn84Rm1cvF/7x+iuGtnwmVpMgVW34UppJgTPLzyUhozlDOLKFMC5uVsAnVlKEtqWJL8FZPXiedq7pn+f11rdEs6ijDGZzDJXhwAw1oQgvawEDBM7zCm2OcF+fd+ViOlpxi5xT+wPn8AbYZkPI=</latexit><latexit sha1_base64="y6UIlSXRZgD9e27DiUCwLf/y8P0=">AAAB8XicbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9oFtKJPppB06mYSZG6GE/oUbF4q49W/c+TdO2iy09cDA4Zx7mXNPkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeLaiFg94CzhfkTHSoSCUbTS4yCiOAnCrDkfVmtu3V2ArBOvIDUo0BpWvwajmKURV8gkNabvuQn6GdUomOTzyiA1PKFsSse8b6miETd+tkg8JxdWGZEw1vYpJAv190ZGI2NmUWAn84Rm1cvF/7x+iuGtnwmVpMgVW34UppJgTPLzyUhozlDOLKFMC5uVsAnVlKEtqWJL8FZPXiedq7pn+f11rdEs6ijDGZzDJXhwAw1oQgvawEDBM7zCm2OcF+fd+ViOlpxi5xT+wPn8AbYZkPI=</latexit><latexit sha1_base64="y6UIlSXRZgD9e27DiUCwLf/y8P0=">AAAB8XicbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9oFtKJPppB06mYSZG6GE/oUbF4q49W/c+TdO2iy09cDA4Zx7mXNPkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeLaiFg94CzhfkTHSoSCUbTS4yCiOAnCrDkfVmtu3V2ArBOvIDUo0BpWvwajmKURV8gkNabvuQn6GdUomOTzyiA1PKFsSse8b6miETd+tkg8JxdWGZEw1vYpJAv190ZGI2NmUWAn84Rm1cvF/7x+iuGtnwmVpMgVW34UppJgTPLzyUhozlDOLKFMC5uVsAnVlKEtqWJL8FZPXiedq7pn+f11rdEs6ijDGZzDJXhwAw1oQgvawEDBM7zCm2OcF+fd+ViOlpxi5xT+wPn8AbYZkPI=</latexit>F<latexit sha1_base64="DjHvM7ny/jrW2XPIMsT/hfgzCGE=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRYE6bKCfWAbymR60w6dTMLMRCihf+HGhSJu/Rt3/o2TNgttPTBwOOde5twTJIJr47rfztr6xubWdmmnvLu3f3BYOTpu6zhVDFssFrHqBlSj4BJbhhuB3UQhjQKBnWBym/udJ1Sax/LBTBP0IzqSPOSMGis99iNqxkGY3c0Glapbc+cgq8QrSBUKNAeVr/4wZmmE0jBBte55bmL8jCrDmcBZuZ9qTCib0BH2LJU0Qu1n88Qzcm6VIQljZZ80ZK7+3shopPU0CuxknlAve7n4n9dLTXjjZ1wmqUHJFh+FqSAmJvn5ZMgVMiOmllCmuM1K2JgqyowtqWxL8JZPXiXty5pn+f1Vtd4o6ijBKZzBBXhwDXVoQBNawEDCM7zCm6OdF+fd+ViMrjnFzgn8gfP5A7MPkPA=</latexit><latexit sha1_base64="DjHvM7ny/jrW2XPIMsT/hfgzCGE=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRYE6bKCfWAbymR60w6dTMLMRCihf+HGhSJu/Rt3/o2TNgttPTBwOOde5twTJIJr47rfztr6xubWdmmnvLu3f3BYOTpu6zhVDFssFrHqBlSj4BJbhhuB3UQhjQKBnWBym/udJ1Sax/LBTBP0IzqSPOSMGis99iNqxkGY3c0Glapbc+cgq8QrSBUKNAeVr/4wZmmE0jBBte55bmL8jCrDmcBZuZ9qTCib0BH2LJU0Qu1n88Qzcm6VIQljZZ80ZK7+3shopPU0CuxknlAve7n4n9dLTXjjZ1wmqUHJFh+FqSAmJvn5ZMgVMiOmllCmuM1K2JgqyowtqWxL8JZPXiXty5pn+f1Vtd4o6ijBKZzBBXhwDXVoQBNawEDCM7zCm6OdF+fd+ViMrjnFzgn8gfP5A7MPkPA=</latexit><latexit sha1_base64="DjHvM7ny/jrW2XPIMsT/hfgzCGE=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRYE6bKCfWAbymR60w6dTMLMRCihf+HGhSJu/Rt3/o2TNgttPTBwOOde5twTJIJr47rfztr6xubWdmmnvLu3f3BYOTpu6zhVDFssFrHqBlSj4BJbhhuB3UQhjQKBnWBym/udJ1Sax/LBTBP0IzqSPOSMGis99iNqxkGY3c0Glapbc+cgq8QrSBUKNAeVr/4wZmmE0jBBte55bmL8jCrDmcBZuZ9qTCib0BH2LJU0Qu1n88Qzcm6VIQljZZ80ZK7+3shopPU0CuxknlAve7n4n9dLTXjjZ1wmqUHJFh+FqSAmJvn5ZMgVMiOmllCmuM1K2JgqyowtqWxL8JZPXiXty5pn+f1Vtd4o6ijBKZzBBXhwDXVoQBNawEDCM7zCm6OdF+fd+ViMrjnFzgn8gfP5A7MPkPA=</latexit><latexit sha1_base64="DjHvM7ny/jrW2XPIMsT/hfgzCGE=">AAAB8XicbVDLSsNAFL3xWeur6tLNYBFclUQEXRYE6bKCfWAbymR60w6dTMLMRCihf+HGhSJu/Rt3/o2TNgttPTBwOOde5twTJIJr47rfztr6xubWdmmnvLu3f3BYOTpu6zhVDFssFrHqBlSj4BJbhhuB3UQhjQKBnWBym/udJ1Sax/LBTBP0IzqSPOSMGis99iNqxkGY3c0Glapbc+cgq8QrSBUKNAeVr/4wZmmE0jBBte55bmL8jCrDmcBZuZ9qTCib0BH2LJU0Qu1n88Qzcm6VIQljZZ80ZK7+3shopPU0CuxknlAve7n4n9dLTXjjZ1wmqUHJFh+FqSAmJvn5ZMgVMiOmllCmuM1K2JgqyowtqWxL8JZPXiXty5pn+f1Vtd4o6ijBKZzBBXhwDXVoQBNawEDCM7zCm6OdF+fd+ViMrjnFzgn8gfP5A7MPkPA=</latexit>

O0

<latexit sha1_base64="7PL7BtpLBdAuEVmh0UrSBv+0Mys=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbcdGcF+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdn8xG1Sqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMknJEq</latexit><latexit sha1_base64="7PL7BtpLBdAuEVmh0UrSBv+0Mys=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbcdGcF+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdn8xG1Sqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMknJEq</latexit><latexit sha1_base64="7PL7BtpLBdAuEVmh0UrSBv+0Mys=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbcdGcF+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdn8xG1Sqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMknJEq</latexit><latexit sha1_base64="7PL7BtpLBdAuEVmh0UrSBv+0Mys=">AAAB8nicbVDLSsNAFL3xWeur6tLNYBFdlUQEXRbcdGcF+4A2lMl00g6dTMLMjVBCP8ONC0Xc+jXu/BsnbRbaemDgcM69zLknSKQw6Lrfztr6xubWdmmnvLu3f3BYOTpumzjVjLdYLGPdDajhUijeQoGSdxPNaRRI3gkmd7nfeeLaiFg94jThfkRHSoSCUbRSrx9RHAdhdn8xG1Sqbs2dg6wSryBVKNAcVL76w5ilEVfIJDWm57kJ+hnVKJjks3I/NTyhbEJHvGepohE3fjaPPCPnVhmSMNb2KSRz9fdGRiNjplFgJ/OIZtnLxf+8XorhrZ8JlaTIFVt8FKaSYEzy+8lQaM5QTi2hTAublbAx1ZShbalsS/CWT14l7auaZ/nDdbXeKOoowSmcwSV4cAN1aEATWsAghmd4hTcHnRfn3flYjK45xc4J/IHz+QMknJEq</latexit>

Q(zr|O)<latexit sha1_base64="FM+7Z3pMnmXZoffo36+hCgcTyLQ=">AAACA3icbVDLSsNAFL3xWesr6k43g0Wom5KIoMuCm+5swT6gDWEynbRDJw9mJkKNBTf+ihsXirj1J9z5N07aCNp6YODMOfdy7z1ezJlUlvVlLC2vrK6tFzaKm1vbO7vm3n5LRokgtEkiHomOhyXlLKRNxRSnnVhQHHictr3RVea3b6mQLApv1DimToAHIfMZwUpLrnnYKPcCrIaen95NXHH/87menLpmyapYU6BFYuekBDnqrvnZ60ckCWioCMdSdm0rVk6KhWKE00mxl0gaYzLCA9rVNMQBlU46vWGCTrTSR34k9AsVmqq/O1IcSDkOPF2ZrSjnvUz8z+smyr90UhbGiaIhmQ3yE45UhLJAUJ8JShQfa4KJYHpXRIZYYKJ0bEUdgj1/8iJpnVVszRvnpWotj6MAR3AMZbDhAqpQgzo0gcADPMELvBqPxrPxZrzPSpeMvOcA/sD4+AZXB5f5</latexit><latexit sha1_base64="FM+7Z3pMnmXZoffo36+hCgcTyLQ=">AAACA3icbVDLSsNAFL3xWesr6k43g0Wom5KIoMuCm+5swT6gDWEynbRDJw9mJkKNBTf+ihsXirj1J9z5N07aCNp6YODMOfdy7z1ezJlUlvVlLC2vrK6tFzaKm1vbO7vm3n5LRokgtEkiHomOhyXlLKRNxRSnnVhQHHictr3RVea3b6mQLApv1DimToAHIfMZwUpLrnnYKPcCrIaen95NXHH/87menLpmyapYU6BFYuekBDnqrvnZ60ckCWioCMdSdm0rVk6KhWKE00mxl0gaYzLCA9rVNMQBlU46vWGCTrTSR34k9AsVmqq/O1IcSDkOPF2ZrSjnvUz8z+smyr90UhbGiaIhmQ3yE45UhLJAUJ8JShQfa4KJYHpXRIZYYKJ0bEUdgj1/8iJpnVVszRvnpWotj6MAR3AMZbDhAqpQgzo0gcADPMELvBqPxrPxZrzPSpeMvOcA/sD4+AZXB5f5</latexit><latexit sha1_base64="FM+7Z3pMnmXZoffo36+hCgcTyLQ=">AAACA3icbVDLSsNAFL3xWesr6k43g0Wom5KIoMuCm+5swT6gDWEynbRDJw9mJkKNBTf+ihsXirj1J9z5N07aCNp6YODMOfdy7z1ezJlUlvVlLC2vrK6tFzaKm1vbO7vm3n5LRokgtEkiHomOhyXlLKRNxRSnnVhQHHictr3RVea3b6mQLApv1DimToAHIfMZwUpLrnnYKPcCrIaen95NXHH/87menLpmyapYU6BFYuekBDnqrvnZ60ckCWioCMdSdm0rVk6KhWKE00mxl0gaYzLCA9rVNMQBlU46vWGCTrTSR34k9AsVmqq/O1IcSDkOPF2ZrSjnvUz8z+smyr90UhbGiaIhmQ3yE45UhLJAUJ8JShQfa4KJYHpXRIZYYKJ0bEUdgj1/8iJpnVVszRvnpWotj6MAR3AMZbDhAqpQgzo0gcADPMELvBqPxrPxZrzPSpeMvOcA/sD4+AZXB5f5</latexit><latexit sha1_base64="FM+7Z3pMnmXZoffo36+hCgcTyLQ=">AAACA3icbVDLSsNAFL3xWesr6k43g0Wom5KIoMuCm+5swT6gDWEynbRDJw9mJkKNBTf+ihsXirj1J9z5N07aCNp6YODMOfdy7z1ezJlUlvVlLC2vrK6tFzaKm1vbO7vm3n5LRokgtEkiHomOhyXlLKRNxRSnnVhQHHictr3RVea3b6mQLApv1DimToAHIfMZwUpLrnnYKPcCrIaen95NXHH/87menLpmyapYU6BFYuekBDnqrvnZ60ckCWioCMdSdm0rVk6KhWKE00mxl0gaYzLCA9rVNMQBlU46vWGCTrTSR34k9AsVmqq/O1IcSDkOPF2ZrSjnvUz8z+smyr90UhbGiaIhmQ3yE45UhLJAUJ8JShQfa4KJYHpXRIZYYKJ0bEUdgj1/8iJpnVVszRvnpWotj6MAR3AMZbDhAqpQgzo0gcADPMELvBqPxrPxZrzPSpeMvOcA/sD4+AZXB5f5</latexit>

N (zs)<latexit sha1_base64="aFI9omamBr3NhJr/u/zMDVmIEgc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIuC266kgr2Am0Ik+mkHTqZhJmJUENx46u4caGIW5/CnW/jJM1CW38Y+PjPOcw5vx8zKpVtfxulldW19Y3yZmVre2d3z9w/6MgoEZi0ccQi0fORJIxy0lZUMdKLBUGhz0jXn1xn9e49EZJG/E5NY+KGaMRpQDFS2vLMo0GI1Bgjlt7Majn7Qfow8+SZZ1btup3LWgangCoUannm12AY4SQkXGGGpOw7dqzcFAlFMSOzyiCRJEZ4gkakr5GjkEg3zU+YWafaGVpBJPTjysrd3xMpCqWchr7uzJaUi7XM/K/WT1Rw5aaUx4kiHM8/ChJmqcjK8rCGVBCs2FQDwoLqXS08RgJhpVOr6BCcxZOXoXNedzTfXlQbzSKOMhzDCdTAgUtoQBNa0AYMj/AMr/BmPBkvxrvxMW8tGcXMIfyR8fkDkLeXig==</latexit><latexit sha1_base64="aFI9omamBr3NhJr/u/zMDVmIEgc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIuC266kgr2Am0Ik+mkHTqZhJmJUENx46u4caGIW5/CnW/jJM1CW38Y+PjPOcw5vx8zKpVtfxulldW19Y3yZmVre2d3z9w/6MgoEZi0ccQi0fORJIxy0lZUMdKLBUGhz0jXn1xn9e49EZJG/E5NY+KGaMRpQDFS2vLMo0GI1Bgjlt7Majn7Qfow8+SZZ1btup3LWgangCoUannm12AY4SQkXGGGpOw7dqzcFAlFMSOzyiCRJEZ4gkakr5GjkEg3zU+YWafaGVpBJPTjysrd3xMpCqWchr7uzJaUi7XM/K/WT1Rw5aaUx4kiHM8/ChJmqcjK8rCGVBCs2FQDwoLqXS08RgJhpVOr6BCcxZOXoXNedzTfXlQbzSKOMhzDCdTAgUtoQBNa0AYMj/AMr/BmPBkvxrvxMW8tGcXMIfyR8fkDkLeXig==</latexit><latexit sha1_base64="aFI9omamBr3NhJr/u/zMDVmIEgc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIuC266kgr2Am0Ik+mkHTqZhJmJUENx46u4caGIW5/CnW/jJM1CW38Y+PjPOcw5vx8zKpVtfxulldW19Y3yZmVre2d3z9w/6MgoEZi0ccQi0fORJIxy0lZUMdKLBUGhz0jXn1xn9e49EZJG/E5NY+KGaMRpQDFS2vLMo0GI1Bgjlt7Majn7Qfow8+SZZ1btup3LWgangCoUannm12AY4SQkXGGGpOw7dqzcFAlFMSOzyiCRJEZ4gkakr5GjkEg3zU+YWafaGVpBJPTjysrd3xMpCqWchr7uzJaUi7XM/K/WT1Rw5aaUx4kiHM8/ChJmqcjK8rCGVBCs2FQDwoLqXS08RgJhpVOr6BCcxZOXoXNedzTfXlQbzSKOMhzDCdTAgUtoQBNa0AYMj/AMr/BmPBkvxrvxMW8tGcXMIfyR8fkDkLeXig==</latexit><latexit sha1_base64="aFI9omamBr3NhJr/u/zMDVmIEgc=">AAACAnicbZDLSsNAFIZP6q3WW9SVuAkWoW5KIoIuC266kgr2Am0Ik+mkHTqZhJmJUENx46u4caGIW5/CnW/jJM1CW38Y+PjPOcw5vx8zKpVtfxulldW19Y3yZmVre2d3z9w/6MgoEZi0ccQi0fORJIxy0lZUMdKLBUGhz0jXn1xn9e49EZJG/E5NY+KGaMRpQDFS2vLMo0GI1Bgjlt7Majn7Qfow8+SZZ1btup3LWgangCoUannm12AY4SQkXGGGpOw7dqzcFAlFMSOzyiCRJEZ4gkakr5GjkEg3zU+YWafaGVpBJPTjysrd3xMpCqWchr7uzJaUi7XM/K/WT1Rw5aaUx4kiHM8/ChJmqcjK8rCGVBCs2FQDwoLqXS08RgJhpVOr6BCcxZOXoXNedzTfXlQbzSKOMhzDCdTAgUtoQBNa0AYMj/AMr/BmPBkvxrvxMW8tGcXMIfyR8fkDkLeXig==</latexit>

z0

s<latexit sha1_base64="0uhKPJ+b+Hqcn8snvWfYmJ3J2b8=">AAAB9HicbVDLSgMxFL3js9ZX1aWbYBFdlRkRdFl047KCfUA7lEyaaUMzmTG5U6hDv8ONC0Xc+jHu/BszbRfaeiBwOOde7skJEikMuu63s7K6tr6xWdgqbu/s7u2XDg4bJk4143UWy1i3Amq4FIrXUaDkrURzGgWSN4Phbe43R1wbEasHHCfcj2hfiVAwilbyOxHFQRBmT5OuOeuWym7FnYIsE29OyjBHrVv66vRilkZcIZPUmLbnJuhnVKNgkk+KndTwhLIh7fO2pYpG3PjZNPSEnFqlR8JY26eQTNXfGxmNjBlHgZ3MQ5pFLxf/89ophtd+JlSSIldsdihMJcGY5A2QntCcoRxbQpkWNithA6opQ9tT0ZbgLX55mTQuKp7l95fl6s28jgIcwwmcgwdXUIU7qEEdGDzCM7zCmzNyXpx352M2uuLMd47gD5zPH/agkjU=</latexit><latexit sha1_base64="0uhKPJ+b+Hqcn8snvWfYmJ3J2b8=">AAAB9HicbVDLSgMxFL3js9ZX1aWbYBFdlRkRdFl047KCfUA7lEyaaUMzmTG5U6hDv8ONC0Xc+jHu/BszbRfaeiBwOOde7skJEikMuu63s7K6tr6xWdgqbu/s7u2XDg4bJk4143UWy1i3Amq4FIrXUaDkrURzGgWSN4Phbe43R1wbEasHHCfcj2hfiVAwilbyOxHFQRBmT5OuOeuWym7FnYIsE29OyjBHrVv66vRilkZcIZPUmLbnJuhnVKNgkk+KndTwhLIh7fO2pYpG3PjZNPSEnFqlR8JY26eQTNXfGxmNjBlHgZ3MQ5pFLxf/89ophtd+JlSSIldsdihMJcGY5A2QntCcoRxbQpkWNithA6opQ9tT0ZbgLX55mTQuKp7l95fl6s28jgIcwwmcgwdXUIU7qEEdGDzCM7zCmzNyXpx352M2uuLMd47gD5zPH/agkjU=</latexit><latexit sha1_base64="0uhKPJ+b+Hqcn8snvWfYmJ3J2b8=">AAAB9HicbVDLSgMxFL3js9ZX1aWbYBFdlRkRdFl047KCfUA7lEyaaUMzmTG5U6hDv8ONC0Xc+jHu/BszbRfaeiBwOOde7skJEikMuu63s7K6tr6xWdgqbu/s7u2XDg4bJk4143UWy1i3Amq4FIrXUaDkrURzGgWSN4Phbe43R1wbEasHHCfcj2hfiVAwilbyOxHFQRBmT5OuOeuWym7FnYIsE29OyjBHrVv66vRilkZcIZPUmLbnJuhnVKNgkk+KndTwhLIh7fO2pYpG3PjZNPSEnFqlR8JY26eQTNXfGxmNjBlHgZ3MQ5pFLxf/89ophtd+JlSSIldsdihMJcGY5A2QntCcoRxbQpkWNithA6opQ9tT0ZbgLX55mTQuKp7l95fl6s28jgIcwwmcgwdXUIU7qEEdGDzCM7zCmzNyXpx352M2uuLMd47gD5zPH/agkjU=</latexit><latexit sha1_base64="0uhKPJ+b+Hqcn8snvWfYmJ3J2b8=">AAAB9HicbVDLSgMxFL3js9ZX1aWbYBFdlRkRdFl047KCfUA7lEyaaUMzmTG5U6hDv8ONC0Xc+jHu/BszbRfaeiBwOOde7skJEikMuu63s7K6tr6xWdgqbu/s7u2XDg4bJk4143UWy1i3Amq4FIrXUaDkrURzGgWSN4Phbe43R1wbEasHHCfcj2hfiVAwilbyOxHFQRBmT5OuOeuWym7FnYIsE29OyjBHrVv66vRilkZcIZPUmLbnJuhnVKNgkk+KndTwhLIh7fO2pYpG3PjZNPSEnFqlR8JY26eQTNXfGxmNjBlHgZ3MQ5pFLxf/89ophtd+JlSSIldsdihMJcGY5A2QntCcoRxbQpkWNithA6opQ9tT0ZbgLX55mTQuKp7l95fl6s28jgIcwwmcgwdXUIU7qEEdGDzCM7zCmzNyXpx352M2uuLMd47gD5zPH/agkjU=</latexit>

I<latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit><latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit><latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit><latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit>

giraffe

giraffe

tree

Figure 2. Overview of our Layout2Im network for generating images from layout during training. The inputs to the model are the ground

truth image with its layout. The objects are first cropped from the input image according to their bounding boxes, and then processed with

the object estimator to predict a latent code for each object. After that, multiple object feature maps are prepared by the object composer

based on the latent codes and layout, and processed with the object encoder, objects fuser and image decoder to reconstruct the input image.

Additional set of latent codes are also sampled from a normal distribution to generate a new image. Finally, objects in generated images

are used to regress the sampled latent codes. The model is trained adversarially against a pair of discriminators and a number of objectives.

appearance-related parts, and only the bounding boxes and

category labels are used during both training and testing.

3. Image Generation from Layout

The overall training pipeline of the proposed approach

is illustrated in Figure 2. Given a ground-truth image I

and its corresponding layout L, where Li = (xi, yi, hi, wi)containing the top-left coordinate, height and width of the

bounding box, our model first samples two latent codes

zri and zsi for each object instance Oi. The zri is sam-

pled from the posterior Q(zr|Oi) conditioned on object Oi

cropped from the input image according to Li. The zsi is

sampled from a normal prior distribution N (zs). Each ob-

ject Oi also has a word embedding wi, which is an em-

bedding of its category label yi. Based on the latent codes

zi ∈ {zri, zsi}, word embedding wi, and layout Li, multi-

ple object feature maps Fi are constructed, and then fed into

the object encoder and the objects fuser sequentially, gener-

ating a fused hidden feature map H containing information

from all specified objects. Finally, an image decoder D is

used to reconstruct, I = D(H), the input ground-truth im-

age I and generate a new image I′, simultaneously; the for-

mer comes from zr = {zri} and the latter from zs = {zsi}.

Notably, both resulting images match the training image in-

put layout. To make the mapping between the generated

object O′

i and the sampled latent code zsi consistent, we

make the object estimator regress the sampled latent codes

zsi based on the generated object O′

i in I′ at locations Li.

To train the model adversarially, we also introduce a pair

of discriminators, Dimg and Dobj, to classify the results at

image and object level as being real or fake.

Once the model is trained, it can generate a new image

from a layout by sampling object latent codes from the nor-

mal prior distribution N (zs) as illustrated in Figure 1.

3.1. Object Latent Code Estimation

Object latent code posterior distributions are first esti-

mated from the ground-truth image, and used to sample ob-

ject latent code zri ∼ Q(zri|Oi) = N (µ(Oi), σ(Oi)).These object latent codes model the ambiguity in object

appearance in the ground-truth image, and play important

roles in reconstructing the input image later.

Figure 3 illustrates the object latent code estimation pro-

cess. First, each object Oi is cropped, from the input image

I according to its bounding box Li, and then resized to fit

the input dimensionality of object estimator using bilinear

interpolation. The resized object crops are fed into an object

estimator which consists of several convolutional layers and

two fully-connected layers. The object estimator predicts

the mean and variance of the posterior distribution for each

input object Oi. Finally, the predicted mean and variance

are used to sample a latent code zri for the input object Oi.

We sample latent code for every object in the input image.

3.2. Object Feature Map Composition

Given the object latent code zi ∈ Rm sampled from ei-

ther posterior or the prior ( zi ∈ {zri, zsi}), object category

label yi and corresponding bounding box information Li,

the object composer module constructs a feature map Fi for

each object Oi. Each feature map Fi contains a region cor-

responding to Li filled with the disentangled representation

of that object, consisting of object identity and appearance.

Figure 4 illustrates this module. The object category la-

bel yi is first transformed to a corresponding word vector

8586

ObjectEstimator

I<latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit><latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit><latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit><latexit sha1_base64="tQM5RlddL0FaWu4rnjk+IbpMa70=">AAAB8XicbVBNS8NAFHypX7V+VT16WSyCp5KIoMeiF71VsLXYhrLZvrRLN5uwuxFK6L/w4kERr/4bb/4bN20O2jqwMMy8x86bIBFcG9f9dkorq2vrG+XNytb2zu5edf+greNUMWyxWMSqE1CNgktsGW4EdhKFNAoEPgTj69x/eEKleSzvzSRBP6JDyUPOqLHSYy+iZhSE2e20X625dXcGsky8gtSgQLNf/eoNYpZGKA0TVOuu5ybGz6gynAmcVnqpxoSyMR1i11JJI9R+Nks8JSdWGZAwVvZJQ2bq742MRlpPosBO5gn1opeL/3nd1ISXfsZlkhqUbP5RmApiYpKfTwZcITNiYgllitushI2ooszYkiq2BG/x5GXSPqt7lt+d1xpXRR1lOIJjOAUPLqABN9CEFjCQ8Ayv8OZo58V5dz7moyWn2DmEP3A+fwC10JDt</latexit>

O1<latexit sha1_base64="eju9Ig8AekyVwYJPkZFOYip32zo=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLoxp0V7AOaUCbTm3boZBJmJkIJ/Q03LhRx68+482+ctllo64GBwzn3cs+cMBVcG9f9dkpr6xubW+Xtys7u3v5B9fCorZNMMWyxRCSqG1KNgktsGW4EdlOFNA4FdsLx7czvPKHSPJGPZpJiENOh5BFn1FjJ92NqRmGU30/7Xr9ac+vuHGSVeAWpQYFmv/rlDxKWxSgNE1TrnuemJsipMpwJnFb8TGNK2ZgOsWeppDHqIJ9nnpIzqwxIlCj7pCFz9fdGTmOtJ3FoJ2cZ9bI3E//zepmJroOcyzQzKNniUJQJYhIyK4AMuEJmxMQSyhS3WQkbUUWZsTVVbAne8pdXSfui7ln+cFlr3BR1lOEETuEcPLiCBtxBE1rAIIVneIU3J3NenHfnYzFacoqdY/gD5/MH67SRlw==</latexit><latexit sha1_base64="eju9Ig8AekyVwYJPkZFOYip32zo=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLoxp0V7AOaUCbTm3boZBJmJkIJ/Q03LhRx68+482+ctllo64GBwzn3cs+cMBVcG9f9dkpr6xubW+Xtys7u3v5B9fCorZNMMWyxRCSqG1KNgktsGW4EdlOFNA4FdsLx7czvPKHSPJGPZpJiENOh5BFn1FjJ92NqRmGU30/7Xr9ac+vuHGSVeAWpQYFmv/rlDxKWxSgNE1TrnuemJsipMpwJnFb8TGNK2ZgOsWeppDHqIJ9nnpIzqwxIlCj7pCFz9fdGTmOtJ3FoJ2cZ9bI3E//zepmJroOcyzQzKNniUJQJYhIyK4AMuEJmxMQSyhS3WQkbUUWZsTVVbAne8pdXSfui7ln+cFlr3BR1lOEETuEcPLiCBtxBE1rAIIVneIU3J3NenHfnYzFacoqdY/gD5/MH67SRlw==</latexit><latexit sha1_base64="eju9Ig8AekyVwYJPkZFOYip32zo=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLoxp0V7AOaUCbTm3boZBJmJkIJ/Q03LhRx68+482+ctllo64GBwzn3cs+cMBVcG9f9dkpr6xubW+Xtys7u3v5B9fCorZNMMWyxRCSqG1KNgktsGW4EdlOFNA4FdsLx7czvPKHSPJGPZpJiENOh5BFn1FjJ92NqRmGU30/7Xr9ac+vuHGSVeAWpQYFmv/rlDxKWxSgNE1TrnuemJsipMpwJnFb8TGNK2ZgOsWeppDHqIJ9nnpIzqwxIlCj7pCFz9fdGTmOtJ3FoJ2cZ9bI3E//zepmJroOcyzQzKNniUJQJYhIyK4AMuEJmxMQSyhS3WQkbUUWZsTVVbAne8pdXSfui7ln+cFlr3BR1lOEETuEcPLiCBtxBE1rAIIVneIU3J3NenHfnYzFacoqdY/gD5/MH67SRlw==</latexit><latexit sha1_base64="eju9Ig8AekyVwYJPkZFOYip32zo=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLoxp0V7AOaUCbTm3boZBJmJkIJ/Q03LhRx68+482+ctllo64GBwzn3cs+cMBVcG9f9dkpr6xubW+Xtys7u3v5B9fCorZNMMWyxRCSqG1KNgktsGW4EdlOFNA4FdsLx7czvPKHSPJGPZpJiENOh5BFn1FjJ92NqRmGU30/7Xr9ac+vuHGSVeAWpQYFmv/rlDxKWxSgNE1TrnuemJsipMpwJnFb8TGNK2ZgOsWeppDHqIJ9nnpIzqwxIlCj7pCFz9fdGTmOtJ3FoJ2cZ9bI3E//zepmJroOcyzQzKNniUJQJYhIyK4AMuEJmxMQSyhS3WQkbUUWZsTVVbAne8pdXSfui7ln+cFlr3BR1lOEETuEcPLiCBtxBE1rAIIVneIU3J3NenHfnYzFacoqdY/gD5/MH67SRlw==</latexit>

O2<latexit sha1_base64="Y89mpVTbtQEHQzTxg9YmMyxQPEs=">AAAB83icbVDLSsNAFL3xWeur6tLNYBFclaQIuiy6cWcF+4CmlMn0ph06mYSZiVBCf8ONC0Xc+jPu/BsnbRbaemDgcM693DMnSATXxnW/nbX1jc2t7dJOeXdv/+CwcnTc1nGqGLZYLGLVDahGwSW2DDcCu4lCGgUCO8HkNvc7T6g0j+WjmSbYj+hI8pAzaqzk+xE14yDM7meD+qBSdWvuHGSVeAWpQoHmoPLlD2OWRigNE1Trnucmpp9RZTgTOCv7qcaEsgkdYc9SSSPU/WyeeUbOrTIkYazsk4bM1d8bGY20nkaBncwz6mUvF//zeqkJr/sZl0lqULLFoTAVxMQkL4AMuUJmxNQSyhS3WQkbU0WZsTWVbQne8pdXSbte8yx/uKw2boo6SnAKZ3ABHlxBA+6gCS1gkMAzvMKbkzovzrvzsRhdc4qdE/gD5/MH7TiRmA==</latexit><latexit sha1_base64="Y89mpVTbtQEHQzTxg9YmMyxQPEs=">AAAB83icbVDLSsNAFL3xWeur6tLNYBFclaQIuiy6cWcF+4CmlMn0ph06mYSZiVBCf8ONC0Xc+jPu/BsnbRbaemDgcM693DMnSATXxnW/nbX1jc2t7dJOeXdv/+CwcnTc1nGqGLZYLGLVDahGwSW2DDcCu4lCGgUCO8HkNvc7T6g0j+WjmSbYj+hI8pAzaqzk+xE14yDM7meD+qBSdWvuHGSVeAWpQoHmoPLlD2OWRigNE1Trnucmpp9RZTgTOCv7qcaEsgkdYc9SSSPU/WyeeUbOrTIkYazsk4bM1d8bGY20nkaBncwz6mUvF//zeqkJr/sZl0lqULLFoTAVxMQkL4AMuUJmxNQSyhS3WQkbU0WZsTWVbQne8pdXSbte8yx/uKw2boo6SnAKZ3ABHlxBA+6gCS1gkMAzvMKbkzovzrvzsRhdc4qdE/gD5/MH7TiRmA==</latexit><latexit sha1_base64="Y89mpVTbtQEHQzTxg9YmMyxQPEs=">AAAB83icbVDLSsNAFL3xWeur6tLNYBFclaQIuiy6cWcF+4CmlMn0ph06mYSZiVBCf8ONC0Xc+jPu/BsnbRbaemDgcM693DMnSATXxnW/nbX1jc2t7dJOeXdv/+CwcnTc1nGqGLZYLGLVDahGwSW2DDcCu4lCGgUCO8HkNvc7T6g0j+WjmSbYj+hI8pAzaqzk+xE14yDM7meD+qBSdWvuHGSVeAWpQoHmoPLlD2OWRigNE1Trnucmpp9RZTgTOCv7qcaEsgkdYc9SSSPU/WyeeUbOrTIkYazsk4bM1d8bGY20nkaBncwz6mUvF//zeqkJr/sZl0lqULLFoTAVxMQkL4AMuUJmxNQSyhS3WQkbU0WZsTWVbQne8pdXSbte8yx/uKw2boo6SnAKZ3ABHlxBA+6gCS1gkMAzvMKbkzovzrvzsRhdc4qdE/gD5/MH7TiRmA==</latexit><latexit sha1_base64="Y89mpVTbtQEHQzTxg9YmMyxQPEs=">AAAB83icbVDLSsNAFL3xWeur6tLNYBFclaQIuiy6cWcF+4CmlMn0ph06mYSZiVBCf8ONC0Xc+jPu/BsnbRbaemDgcM693DMnSATXxnW/nbX1jc2t7dJOeXdv/+CwcnTc1nGqGLZYLGLVDahGwSW2DDcCu4lCGgUCO8HkNvc7T6g0j+WjmSbYj+hI8pAzaqzk+xE14yDM7meD+qBSdWvuHGSVeAWpQoHmoPLlD2OWRigNE1Trnucmpp9RZTgTOCv7qcaEsgkdYc9SSSPU/WyeeUbOrTIkYazsk4bM1d8bGY20nkaBncwz6mUvF//zeqkJr/sZl0lqULLFoTAVxMQkL4AMuUJmxNQSyhS3WQkbU0WZsTWVbQne8pdXSbte8yx/uKw2boo6SnAKZ3ABHlxBA+6gCS1gkMAzvMKbkzovzrvzsRhdc4qdE/gD5/MH7TiRmA==</latexit>

O3<latexit sha1_base64="71vlDT9My06Wqc9YxtVH44adNgc=">AAAB83icbVBNS8NAFHypX7V+VT16WSyCp5KooMeiF29WsLbQlLLZbtqlm03YfRFK6N/w4kERr/4Zb/4bN20O2jqwMMy8x5udIJHCoOt+O6WV1bX1jfJmZWt7Z3evun/waOJUM95isYx1J6CGS6F4CwVK3kk0p1EgeTsY3+R++4lrI2L1gJOE9yI6VCIUjKKVfD+iOArC7G7aP+9Xa27dnYEsE68gNSjQ7Fe//EHM0ogrZJIa0/XcBHsZ1SiY5NOKnxqeUDamQ961VNGIm142yzwlJ1YZkDDW9ikkM/X3RkYjYyZRYCfzjGbRy8X/vG6K4VUvEypJkSs2PxSmkmBM8gLIQGjOUE4soUwLm5WwEdWUoa2pYkvwFr+8TB7P6p7l9xe1xnVRRxmO4BhOwYNLaMAtNKEFDBJ4hld4c1LnxXl3PuajJafYOYQ/cD5/AO68kZk=</latexit><latexit sha1_base64="71vlDT9My06Wqc9YxtVH44adNgc=">AAAB83icbVBNS8NAFHypX7V+VT16WSyCp5KooMeiF29WsLbQlLLZbtqlm03YfRFK6N/w4kERr/4Zb/4bN20O2jqwMMy8x5udIJHCoOt+O6WV1bX1jfJmZWt7Z3evun/waOJUM95isYx1J6CGS6F4CwVK3kk0p1EgeTsY3+R++4lrI2L1gJOE9yI6VCIUjKKVfD+iOArC7G7aP+9Xa27dnYEsE68gNSjQ7Fe//EHM0ogrZJIa0/XcBHsZ1SiY5NOKnxqeUDamQ961VNGIm142yzwlJ1YZkDDW9ikkM/X3RkYjYyZRYCfzjGbRy8X/vG6K4VUvEypJkSs2PxSmkmBM8gLIQGjOUE4soUwLm5WwEdWUoa2pYkvwFr+8TB7P6p7l9xe1xnVRRxmO4BhOwYNLaMAtNKEFDBJ4hld4c1LnxXl3PuajJafYOYQ/cD5/AO68kZk=</latexit><latexit sha1_base64="71vlDT9My06Wqc9YxtVH44adNgc=">AAAB83icbVBNS8NAFHypX7V+VT16WSyCp5KooMeiF29WsLbQlLLZbtqlm03YfRFK6N/w4kERr/4Zb/4bN20O2jqwMMy8x5udIJHCoOt+O6WV1bX1jfJmZWt7Z3evun/waOJUM95isYx1J6CGS6F4CwVK3kk0p1EgeTsY3+R++4lrI2L1gJOE9yI6VCIUjKKVfD+iOArC7G7aP+9Xa27dnYEsE68gNSjQ7Fe//EHM0ogrZJIa0/XcBHsZ1SiY5NOKnxqeUDamQ961VNGIm142yzwlJ1YZkDDW9ikkM/X3RkYjYyZRYCfzjGbRy8X/vG6K4VUvEypJkSs2PxSmkmBM8gLIQGjOUE4soUwLm5WwEdWUoa2pYkvwFr+8TB7P6p7l9xe1xnVRRxmO4BhOwYNLaMAtNKEFDBJ4hld4c1LnxXl3PuajJafYOYQ/cD5/AO68kZk=</latexit><latexit sha1_base64="71vlDT9My06Wqc9YxtVH44adNgc=">AAAB83icbVBNS8NAFHypX7V+VT16WSyCp5KooMeiF29WsLbQlLLZbtqlm03YfRFK6N/w4kERr/4Zb/4bN20O2jqwMMy8x5udIJHCoOt+O6WV1bX1jfJmZWt7Z3evun/waOJUM95isYx1J6CGS6F4CwVK3kk0p1EgeTsY3+R++4lrI2L1gJOE9yI6VCIUjKKVfD+iOArC7G7aP+9Xa27dnYEsE68gNSjQ7Fe//EHM0ogrZJIa0/XcBHsZ1SiY5NOKnxqeUDamQ961VNGIm142yzwlJ1YZkDDW9ikkM/X3RkYjYyZRYCfzjGbRy8X/vG6K4VUvEypJkSs2PxSmkmBM8gLIQGjOUE4soUwLm5WwEdWUoa2pYkvwFr+8TB7P6p7l9xe1xnVRRxmO4BhOwYNLaMAtNKEFDBJ4hld4c1LnxXl3PuajJafYOYQ/cD5/AO68kZk=</latexit>

Q(zr1|O1)<latexit sha1_base64="zw4K/UOt9PVi/6RTYSCH1TLNDfw=">AAACCHicbVDLSsNAFL3xWesr6tKFg0Wom5KIoMuiG3e2YB/QljCZTtqhkwczE6HGLN34K25cKOLWT3Dn3zhpI2jrgYEz59zLvfe4EWdSWdaXsbC4tLyyWlgrrm9sbm2bO7tNGcaC0AYJeSjaLpaUs4A2FFOctiNBse9y2nJHl5nfuqVCsjC4UeOI9nw8CJjHCFZacsyDernrYzV0veQudRJhp/c//+vUsY8ds2RVrAnQPLFzUoIcNcf87PZDEvs0UIRjKTu2FalegoVihNO02I0ljTAZ4QHtaBpgn8peMjkkRUda6SMvFPoFCk3U3x0J9qUc+66uzJaUs14m/ud1YuWd9xIWRLGiAZkO8mKOVIiyVFCfCUoUH2uCiWB6V0SGWGCidHZFHYI9e/I8aZ5UbM3rp6XqRR5HAfbhEMpgwxlU4Qpq0AACD/AEL/BqPBrPxpvxPi1dMPKePfgD4+Mb4xiZ3g==</latexit><latexit sha1_base64="zw4K/UOt9PVi/6RTYSCH1TLNDfw=">AAACCHicbVDLSsNAFL3xWesr6tKFg0Wom5KIoMuiG3e2YB/QljCZTtqhkwczE6HGLN34K25cKOLWT3Dn3zhpI2jrgYEz59zLvfe4EWdSWdaXsbC4tLyyWlgrrm9sbm2bO7tNGcaC0AYJeSjaLpaUs4A2FFOctiNBse9y2nJHl5nfuqVCsjC4UeOI9nw8CJjHCFZacsyDernrYzV0veQudRJhp/c//+vUsY8ds2RVrAnQPLFzUoIcNcf87PZDEvs0UIRjKTu2FalegoVihNO02I0ljTAZ4QHtaBpgn8peMjkkRUda6SMvFPoFCk3U3x0J9qUc+66uzJaUs14m/ud1YuWd9xIWRLGiAZkO8mKOVIiyVFCfCUoUH2uCiWB6V0SGWGCidHZFHYI9e/I8aZ5UbM3rp6XqRR5HAfbhEMpgwxlU4Qpq0AACD/AEL/BqPBrPxpvxPi1dMPKePfgD4+Mb4xiZ3g==</latexit><latexit sha1_base64="zw4K/UOt9PVi/6RTYSCH1TLNDfw=">AAACCHicbVDLSsNAFL3xWesr6tKFg0Wom5KIoMuiG3e2YB/QljCZTtqhkwczE6HGLN34K25cKOLWT3Dn3zhpI2jrgYEz59zLvfe4EWdSWdaXsbC4tLyyWlgrrm9sbm2bO7tNGcaC0AYJeSjaLpaUs4A2FFOctiNBse9y2nJHl5nfuqVCsjC4UeOI9nw8CJjHCFZacsyDernrYzV0veQudRJhp/c//+vUsY8ds2RVrAnQPLFzUoIcNcf87PZDEvs0UIRjKTu2FalegoVihNO02I0ljTAZ4QHtaBpgn8peMjkkRUda6SMvFPoFCk3U3x0J9qUc+66uzJaUs14m/ud1YuWd9xIWRLGiAZkO8mKOVIiyVFCfCUoUH2uCiWB6V0SGWGCidHZFHYI9e/I8aZ5UbM3rp6XqRR5HAfbhEMpgwxlU4Qpq0AACD/AEL/BqPBrPxpvxPi1dMPKePfgD4+Mb4xiZ3g==</latexit><latexit sha1_base64="zw4K/UOt9PVi/6RTYSCH1TLNDfw=">AAACCHicbVDLSsNAFL3xWesr6tKFg0Wom5KIoMuiG3e2YB/QljCZTtqhkwczE6HGLN34K25cKOLWT3Dn3zhpI2jrgYEz59zLvfe4EWdSWdaXsbC4tLyyWlgrrm9sbm2bO7tNGcaC0AYJeSjaLpaUs4A2FFOctiNBse9y2nJHl5nfuqVCsjC4UeOI9nw8CJjHCFZacsyDernrYzV0veQudRJhp/c//+vUsY8ds2RVrAnQPLFzUoIcNcf87PZDEvs0UIRjKTu2FalegoVihNO02I0ljTAZ4QHtaBpgn8peMjkkRUda6SMvFPoFCk3U3x0J9qUc+66uzJaUs14m/ud1YuWd9xIWRLGiAZkO8mKOVIiyVFCfCUoUH2uCiWB6V0SGWGCidHZFHYI9e/I8aZ5UbM3rp6XqRR5HAfbhEMpgwxlU4Qpq0AACD/AEL/BqPBrPxpvxPi1dMPKePfgD4+Mb4xiZ3g==</latexit>

Q(zr2|O2)<latexit sha1_base64="9P4+CWVIK8iqf/ZZqqX1ZEp3KNc=">AAACCHicbVDLSsNAFJ3UV62vqEsXDhahbkpSBF0W3bizBfuANoTJdNIOnUzCzESoMUs3/oobF4q49RPc+TdO2gjaemDgzDn3cu89XsSoVJb1ZRSWlldW14rrpY3Nre0dc3evLcNYYNLCIQtF10OSMMpJS1HFSDcSBAUeIx1vfJn5nVsiJA35jZpExAnQkFOfYqS05JqHzUo/QGrk+cld6iailt7//K9Tt3bimmWrak0BF4mdkzLI0XDNz/4gxHFAuMIMSdmzrUg5CRKKYkbSUj+WJEJ4jIakpylHAZFOMj0khcdaGUA/FPpxBafq744EBVJOAk9XZkvKeS8T//N6sfLPnYTyKFaE49kgP2ZQhTBLBQ6oIFixiSYIC6p3hXiEBMJKZ1fSIdjzJy+Sdq1qa948Ldcv8jiK4AAcgQqwwRmogyvQAC2AwQN4Ai/g1Xg0no03431WWjDynn3wB8bHN+YwmeA=</latexit><latexit sha1_base64="9P4+CWVIK8iqf/ZZqqX1ZEp3KNc=">AAACCHicbVDLSsNAFJ3UV62vqEsXDhahbkpSBF0W3bizBfuANoTJdNIOnUzCzESoMUs3/oobF4q49RPc+TdO2gjaemDgzDn3cu89XsSoVJb1ZRSWlldW14rrpY3Nre0dc3evLcNYYNLCIQtF10OSMMpJS1HFSDcSBAUeIx1vfJn5nVsiJA35jZpExAnQkFOfYqS05JqHzUo/QGrk+cld6iailt7//K9Tt3bimmWrak0BF4mdkzLI0XDNz/4gxHFAuMIMSdmzrUg5CRKKYkbSUj+WJEJ4jIakpylHAZFOMj0khcdaGUA/FPpxBafq744EBVJOAk9XZkvKeS8T//N6sfLPnYTyKFaE49kgP2ZQhTBLBQ6oIFixiSYIC6p3hXiEBMJKZ1fSIdjzJy+Sdq1qa948Ldcv8jiK4AAcgQqwwRmogyvQAC2AwQN4Ai/g1Xg0no03431WWjDynn3wB8bHN+YwmeA=</latexit><latexit sha1_base64="9P4+CWVIK8iqf/ZZqqX1ZEp3KNc=">AAACCHicbVDLSsNAFJ3UV62vqEsXDhahbkpSBF0W3bizBfuANoTJdNIOnUzCzESoMUs3/oobF4q49RPc+TdO2gjaemDgzDn3cu89XsSoVJb1ZRSWlldW14rrpY3Nre0dc3evLcNYYNLCIQtF10OSMMpJS1HFSDcSBAUeIx1vfJn5nVsiJA35jZpExAnQkFOfYqS05JqHzUo/QGrk+cld6iailt7//K9Tt3bimmWrak0BF4mdkzLI0XDNz/4gxHFAuMIMSdmzrUg5CRKKYkbSUj+WJEJ4jIakpylHAZFOMj0khcdaGUA/FPpxBafq744EBVJOAk9XZkvKeS8T//N6sfLPnYTyKFaE49kgP2ZQhTBLBQ6oIFixiSYIC6p3hXiEBMJKZ1fSIdjzJy+Sdq1qa948Ldcv8jiK4AAcgQqwwRmogyvQAC2AwQN4Ai/g1Xg0no03431WWjDynn3wB8bHN+YwmeA=</latexit><latexit sha1_base64="9P4+CWVIK8iqf/ZZqqX1ZEp3KNc=">AAACCHicbVDLSsNAFJ3UV62vqEsXDhahbkpSBF0W3bizBfuANoTJdNIOnUzCzESoMUs3/oobF4q49RPc+TdO2gjaemDgzDn3cu89XsSoVJb1ZRSWlldW14rrpY3Nre0dc3evLcNYYNLCIQtF10OSMMpJS1HFSDcSBAUeIx1vfJn5nVsiJA35jZpExAnQkFOfYqS05JqHzUo/QGrk+cld6iailt7//K9Tt3bimmWrak0BF4mdkzLI0XDNz/4gxHFAuMIMSdmzrUg5CRKKYkbSUj+WJEJ4jIakpylHAZFOMj0khcdaGUA/FPpxBafq744EBVJOAk9XZkvKeS8T//N6sfLPnYTyKFaE49kgP2ZQhTBLBQ6oIFixiSYIC6p3hXiEBMJKZ1fSIdjzJy+Sdq1qa948Ldcv8jiK4AAcgQqwwRmogyvQAC2AwQN4Ai/g1Xg0no03431WWjDynn3wB8bHN+YwmeA=</latexit>

Q(zr3|O3)<latexit sha1_base64="Pccmin44nAzm1cDaNQPLeNd24Sk=">AAACCHicbVDLSsNAFJ34rPUVdenCwSLUTUmsoMuiG3e2YB/QhjCZTtqhk0mYmQg1ZunGX3HjQhG3foI7/8ZJG0FbDwycOede7r3HixiVyrK+jIXFpeWV1cJacX1jc2vb3NltyTAWmDRxyELR8ZAkjHLSVFQx0okEQYHHSNsbXWZ++5YISUN+o8YRcQI04NSnGCktueZBo9wLkBp6fnKXuomopvc//+vUrR67ZsmqWBPAeWLnpARy1F3zs9cPcRwQrjBDUnZtK1JOgoSimJG02IsliRAeoQHpaspRQKSTTA5J4ZFW+tAPhX5cwYn6uyNBgZTjwNOV2ZJy1svE/7xurPxzJ6E8ihXheDrIjxlUIcxSgX0qCFZsrAnCgupdIR4igbDS2RV1CPbsyfOkdVKxNW+clmoXeRwFsA8OQRnY4AzUwBWogybA4AE8gRfwajwaz8ab8T4tXTDynj3wB8bHN+lImeI=</latexit><latexit sha1_base64="Pccmin44nAzm1cDaNQPLeNd24Sk=">AAACCHicbVDLSsNAFJ34rPUVdenCwSLUTUmsoMuiG3e2YB/QhjCZTtqhk0mYmQg1ZunGX3HjQhG3foI7/8ZJG0FbDwycOede7r3HixiVyrK+jIXFpeWV1cJacX1jc2vb3NltyTAWmDRxyELR8ZAkjHLSVFQx0okEQYHHSNsbXWZ++5YISUN+o8YRcQI04NSnGCktueZBo9wLkBp6fnKXuomopvc//+vUrR67ZsmqWBPAeWLnpARy1F3zs9cPcRwQrjBDUnZtK1JOgoSimJG02IsliRAeoQHpaspRQKSTTA5J4ZFW+tAPhX5cwYn6uyNBgZTjwNOV2ZJy1svE/7xurPxzJ6E8ihXheDrIjxlUIcxSgX0qCFZsrAnCgupdIR4igbDS2RV1CPbsyfOkdVKxNW+clmoXeRwFsA8OQRnY4AzUwBWogybA4AE8gRfwajwaz8ab8T4tXTDynj3wB8bHN+lImeI=</latexit><latexit sha1_base64="Pccmin44nAzm1cDaNQPLeNd24Sk=">AAACCHicbVDLSsNAFJ34rPUVdenCwSLUTUmsoMuiG3e2YB/QhjCZTtqhk0mYmQg1ZunGX3HjQhG3foI7/8ZJG0FbDwycOede7r3HixiVyrK+jIXFpeWV1cJacX1jc2vb3NltyTAWmDRxyELR8ZAkjHLSVFQx0okEQYHHSNsbXWZ++5YISUN+o8YRcQI04NSnGCktueZBo9wLkBp6fnKXuomopvc//+vUrR67ZsmqWBPAeWLnpARy1F3zs9cPcRwQrjBDUnZtK1JOgoSimJG02IsliRAeoQHpaspRQKSTTA5J4ZFW+tAPhX5cwYn6uyNBgZTjwNOV2ZJy1svE/7xurPxzJ6E8ihXheDrIjxlUIcxSgX0qCFZsrAnCgupdIR4igbDS2RV1CPbsyfOkdVKxNW+clmoXeRwFsA8OQRnY4AzUwBWogybA4AE8gRfwajwaz8ab8T4tXTDynj3wB8bHN+lImeI=</latexit><latexit sha1_base64="Pccmin44nAzm1cDaNQPLeNd24Sk=">AAACCHicbVDLSsNAFJ34rPUVdenCwSLUTUmsoMuiG3e2YB/QhjCZTtqhk0mYmQg1ZunGX3HjQhG3foI7/8ZJG0FbDwycOede7r3HixiVyrK+jIXFpeWV1cJacX1jc2vb3NltyTAWmDRxyELR8ZAkjHLSVFQx0okEQYHHSNsbXWZ++5YISUN+o8YRcQI04NSnGCktueZBo9wLkBp6fnKXuomopvc//+vUrR67ZsmqWBPAeWLnpARy1F3zs9cPcRwQrjBDUnZtK1JOgoSimJG02IsliRAeoQHpaspRQKSTTA5J4ZFW+tAPhX5cwYn6uyNBgZTjwNOV2ZJy1svE/7xurPxzJ6E8ihXheDrIjxlUIcxSgX0qCFZsrAnCgupdIR4igbDS2RV1CPbsyfOkdVKxNW+clmoXeRwFsA8OQRnY4AzUwBWogybA4AE8gRfwajwaz8ab8T4tXTDynj3wB8bHN+lImeI=</latexit>

zr1<latexit sha1_base64="qIcwpt2S3cVf2P+AOOP616oO7J8=">AAAB+HicbVDLSsNAFL2pr1ofjbp0M1gEVyURQZdFNy4r2Ae0IUymk3boZBJmJkIb8iVuXCji1k9x5984abPQ1gMDh3Pu5Z45QcKZ0o7zbVU2Nre2d6q7tb39g8O6fXTcVXEqCe2QmMeyH2BFORO0o5nmtJ9IiqOA014wvSv83hOVisXiUc8S6kV4LFjICNZG8u36MMJ6EoTZPPcz6ea+3XCazgJonbglaUCJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJrXhqmiCSZTPKYDQwWOqPKyRfAcnRtlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTMluKtfXifdy6Zr+MNVo3Vb1lGFUziDC3DhGlpwD23oAIEUnuEV3qy59WK9Wx/L0YpV7pzAH1ifP0lRk3s=</latexit><latexit sha1_base64="qIcwpt2S3cVf2P+AOOP616oO7J8=">AAAB+HicbVDLSsNAFL2pr1ofjbp0M1gEVyURQZdFNy4r2Ae0IUymk3boZBJmJkIb8iVuXCji1k9x5984abPQ1gMDh3Pu5Z45QcKZ0o7zbVU2Nre2d6q7tb39g8O6fXTcVXEqCe2QmMeyH2BFORO0o5nmtJ9IiqOA014wvSv83hOVisXiUc8S6kV4LFjICNZG8u36MMJ6EoTZPPcz6ea+3XCazgJonbglaUCJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJrXhqmiCSZTPKYDQwWOqPKyRfAcnRtlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTMluKtfXifdy6Zr+MNVo3Vb1lGFUziDC3DhGlpwD23oAIEUnuEV3qy59WK9Wx/L0YpV7pzAH1ifP0lRk3s=</latexit><latexit sha1_base64="qIcwpt2S3cVf2P+AOOP616oO7J8=">AAAB+HicbVDLSsNAFL2pr1ofjbp0M1gEVyURQZdFNy4r2Ae0IUymk3boZBJmJkIb8iVuXCji1k9x5984abPQ1gMDh3Pu5Z45QcKZ0o7zbVU2Nre2d6q7tb39g8O6fXTcVXEqCe2QmMeyH2BFORO0o5nmtJ9IiqOA014wvSv83hOVisXiUc8S6kV4LFjICNZG8u36MMJ6EoTZPPcz6ea+3XCazgJonbglaUCJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJrXhqmiCSZTPKYDQwWOqPKyRfAcnRtlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTMluKtfXifdy6Zr+MNVo3Vb1lGFUziDC3DhGlpwD23oAIEUnuEV3qy59WK9Wx/L0YpV7pzAH1ifP0lRk3s=</latexit><latexit sha1_base64="qIcwpt2S3cVf2P+AOOP616oO7J8=">AAAB+HicbVDLSsNAFL2pr1ofjbp0M1gEVyURQZdFNy4r2Ae0IUymk3boZBJmJkIb8iVuXCji1k9x5984abPQ1gMDh3Pu5Z45QcKZ0o7zbVU2Nre2d6q7tb39g8O6fXTcVXEqCe2QmMeyH2BFORO0o5nmtJ9IiqOA014wvSv83hOVisXiUc8S6kV4LFjICNZG8u36MMJ6EoTZPPcz6ea+3XCazgJonbglaUCJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJrXhqmiCSZTPKYDQwWOqPKyRfAcnRtlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTMluKtfXifdy6Zr+MNVo3Vb1lGFUziDC3DhGlpwD23oAIEUnuEV3qy59WK9Wx/L0YpV7pzAH1ifP0lRk3s=</latexit>

zr2<latexit sha1_base64="FUFMGpEnl/j+JJBe2o64+c/8ZF8=">AAAB+HicbVDLSsNAFL3xWeujUZduBovgqiRF0GXRjcsK9gFtCJPppB06mYSZidCGfIkbF4q49VPc+TdO2iy09cDA4Zx7uWdOkHCmtON8WxubW9s7u5W96v7B4VHNPj7pqjiVhHZIzGPZD7CinAna0Uxz2k8kxVHAaS+Y3hV+74lKxWLxqGcJ9SI8FixkBGsj+XZtGGE9CcJsnvuZbOa+XXcazgJonbglqUOJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJpXh6miCSZTPKYDQwWOqPKyRfAcXRhlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTUluKtfXifdZsM1/OGq3rot66jAGZzDJbhwDS24hzZ0gEAKz/AKb9bcerHerY/l6IZV7pzCH1ifP0rWk3w=</latexit><latexit sha1_base64="FUFMGpEnl/j+JJBe2o64+c/8ZF8=">AAAB+HicbVDLSsNAFL3xWeujUZduBovgqiRF0GXRjcsK9gFtCJPppB06mYSZidCGfIkbF4q49VPc+TdO2iy09cDA4Zx7uWdOkHCmtON8WxubW9s7u5W96v7B4VHNPj7pqjiVhHZIzGPZD7CinAna0Uxz2k8kxVHAaS+Y3hV+74lKxWLxqGcJ9SI8FixkBGsj+XZtGGE9CcJsnvuZbOa+XXcazgJonbglqUOJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJpXh6miCSZTPKYDQwWOqPKyRfAcXRhlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTUluKtfXifdZsM1/OGq3rot66jAGZzDJbhwDS24hzZ0gEAKz/AKb9bcerHerY/l6IZV7pzCH1ifP0rWk3w=</latexit><latexit sha1_base64="FUFMGpEnl/j+JJBe2o64+c/8ZF8=">AAAB+HicbVDLSsNAFL3xWeujUZduBovgqiRF0GXRjcsK9gFtCJPppB06mYSZidCGfIkbF4q49VPc+TdO2iy09cDA4Zx7uWdOkHCmtON8WxubW9s7u5W96v7B4VHNPj7pqjiVhHZIzGPZD7CinAna0Uxz2k8kxVHAaS+Y3hV+74lKxWLxqGcJ9SI8FixkBGsj+XZtGGE9CcJsnvuZbOa+XXcazgJonbglqUOJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJpXh6miCSZTPKYDQwWOqPKyRfAcXRhlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTUluKtfXifdZsM1/OGq3rot66jAGZzDJbhwDS24hzZ0gEAKz/AKb9bcerHerY/l6IZV7pzCH1ifP0rWk3w=</latexit><latexit sha1_base64="FUFMGpEnl/j+JJBe2o64+c/8ZF8=">AAAB+HicbVDLSsNAFL3xWeujUZduBovgqiRF0GXRjcsK9gFtCJPppB06mYSZidCGfIkbF4q49VPc+TdO2iy09cDA4Zx7uWdOkHCmtON8WxubW9s7u5W96v7B4VHNPj7pqjiVhHZIzGPZD7CinAna0Uxz2k8kxVHAaS+Y3hV+74lKxWLxqGcJ9SI8FixkBGsj+XZtGGE9CcJsnvuZbOa+XXcazgJonbglqUOJtm9/DUcxSSMqNOFYqYHrJNrLsNSMcJpXh6miCSZTPKYDQwWOqPKyRfAcXRhlhMJYmic0Wqi/NzIcKTWLAjNZxFSrXiH+5w1SHd54GRNJqqkgy0NhypGOUdECGjFJieYzQzCRzGRFZIIlJtp0VTUluKtfXifdZsM1/OGq3rot66jAGZzDJbhwDS24hzZ0gEAKz/AKb9bcerHerY/l6IZV7pzCH1ifP0rWk3w=</latexit>

zr3<latexit sha1_base64="//DejuoxAqdMcU0GSLo+24SD6Wk=">AAAB+HicbVBNS8NAFHypX7V+NOrRy2IRPJVEBT0WvXisYGuhDWGz3bRLN5uwuxHakF/ixYMiXv0p3vw3btoctHVgYZh5jzc7QcKZ0o7zbVXW1jc2t6rbtZ3dvf26fXDYVXEqCe2QmMeyF2BFORO0o5nmtJdIiqOA08dgclv4j09UKhaLBz1NqBfhkWAhI1gbybfrgwjrcRBms9zP5EXu2w2n6cyBVolbkgaUaPv212AYkzSiQhOOleq7TqK9DEvNCKd5bZAqmmAywSPaN1TgiCovmwfP0alRhiiMpXlCo7n6eyPDkVLTKDCTRUy17BXif14/1eG1lzGRpJoKsjgUphzpGBUtoCGTlGg+NQQTyUxWRMZYYqJNVzVTgrv85VXSPW+6ht9fNlo3ZR1VOIYTOAMXrqAFd9CGDhBI4Rle4c2aWS/Wu/WxGK1Y5c4R/IH1+QNMW5N9</latexit><latexit sha1_base64="//DejuoxAqdMcU0GSLo+24SD6Wk=">AAAB+HicbVBNS8NAFHypX7V+NOrRy2IRPJVEBT0WvXisYGuhDWGz3bRLN5uwuxHakF/ixYMiXv0p3vw3btoctHVgYZh5jzc7QcKZ0o7zbVXW1jc2t6rbtZ3dvf26fXDYVXEqCe2QmMeyF2BFORO0o5nmtJdIiqOA08dgclv4j09UKhaLBz1NqBfhkWAhI1gbybfrgwjrcRBms9zP5EXu2w2n6cyBVolbkgaUaPv212AYkzSiQhOOleq7TqK9DEvNCKd5bZAqmmAywSPaN1TgiCovmwfP0alRhiiMpXlCo7n6eyPDkVLTKDCTRUy17BXif14/1eG1lzGRpJoKsjgUphzpGBUtoCGTlGg+NQQTyUxWRMZYYqJNVzVTgrv85VXSPW+6ht9fNlo3ZR1VOIYTOAMXrqAFd9CGDhBI4Rle4c2aWS/Wu/WxGK1Y5c4R/IH1+QNMW5N9</latexit><latexit sha1_base64="//DejuoxAqdMcU0GSLo+24SD6Wk=">AAAB+HicbVBNS8NAFHypX7V+NOrRy2IRPJVEBT0WvXisYGuhDWGz3bRLN5uwuxHakF/ixYMiXv0p3vw3btoctHVgYZh5jzc7QcKZ0o7zbVXW1jc2t6rbtZ3dvf26fXDYVXEqCe2QmMeyF2BFORO0o5nmtJdIiqOA08dgclv4j09UKhaLBz1NqBfhkWAhI1gbybfrgwjrcRBms9zP5EXu2w2n6cyBVolbkgaUaPv212AYkzSiQhOOleq7TqK9DEvNCKd5bZAqmmAywSPaN1TgiCovmwfP0alRhiiMpXlCo7n6eyPDkVLTKDCTRUy17BXif14/1eG1lzGRpJoKsjgUphzpGBUtoCGTlGg+NQQTyUxWRMZYYqJNVzVTgrv85VXSPW+6ht9fNlo3ZR1VOIYTOAMXrqAFd9CGDhBI4Rle4c2aWS/Wu/WxGK1Y5c4R/IH1+QNMW5N9</latexit><latexit sha1_base64="//DejuoxAqdMcU0GSLo+24SD6Wk=">AAAB+HicbVBNS8NAFHypX7V+NOrRy2IRPJVEBT0WvXisYGuhDWGz3bRLN5uwuxHakF/ixYMiXv0p3vw3btoctHVgYZh5jzc7QcKZ0o7zbVXW1jc2t6rbtZ3dvf26fXDYVXEqCe2QmMeyF2BFORO0o5nmtJdIiqOA08dgclv4j09UKhaLBz1NqBfhkWAhI1gbybfrgwjrcRBms9zP5EXu2w2n6cyBVolbkgaUaPv212AYkzSiQhOOleq7TqK9DEvNCKd5bZAqmmAywSPaN1TgiCovmwfP0alRhiiMpXlCo7n6eyPDkVLTKDCTRUy17BXif14/1eG1lzGRpJoKsjgUphzpGBUtoCGTlGg+NQQTyUxWRMZYYqJNVzVTgrv85VXSPW+6ht9fNlo3ZR1VOIYTOAMXrqAFd9CGDhBI4Rle4c2aWS/Wu/WxGK1Y5c4R/IH1+QNMW5N9</latexit>

Figure 3. Object latent code estimation. Given the input image

and its layout, the objects are first cropped and resized from the

input image. Then the object estimator predicts a distribution for

each object from the object crops, and multiple latent codes are

sampled from the estimated distribution.

embedding wi ∈ Rn, and then concatenated with the ob-

ject latent vector zi. This results in the representation of

the object which has two parts: object embedding and ob-

ject latent code. Intuitively, the object embedding encodes

the identity of the object, while the latent code encodes the

appearance of a specific instance of that object. Jointly

these two components encode sufficient information to re-

construct a specific instance of the object in an image. The

object feature map Fi is composed by simply filling the re-

gion within its bounding box with this object representation

(wi, zi) ∈ Rm+n. For each tuple < yi, zi,Li > encod-

ing object label, latent code and bounding box, we compose

an object feature map Fi. These object feature maps are

downsampled by an object encoder network which contains

several convolutional layers. Then an object fuser module

is used to fuse all the downsampled object feature maps,

generating a hidden feature map H.

3.3. Object Feature Maps Fusion

Since the result image will be decoded from it, a good

hidden feature map H is crucial to generating a realistic

image. The properties of a good hidden feature map can

be summarized as follows: (i) it should encode all object

instances in the desired locations; (ii) it should coordinate

object representations based on other objects in the image;

(iii) it should be able to fill the unspecified regions, e.g.,

background, by implicitly reasoning about plausibility of

the scene with respect to the specified objects.

To satisfy these requirements, we choose a multi-layer

convolutional Long-Short-Term Memory (cLSTM) net-

work [38] to fuse the downsampled object feature maps F.

Different from the traditional LSTM [10], the hidden states

and cell states in cLSTM are both feature maps rather than

vectors. The computation of different gates are also done

by convolutional layers. Therefore, cLSTM can better pre-

serve the spatial information compared with the traditional

vector-based LSTM. The cLSTM acts like an encoder to

Word

Embedding“giraffe”

… … …z1

<latexit sha1_base64="O4eplVAdWsWi2nj78Fp1SNWOEso=">AAAB9XicbVDLSgMxFL2pr1pfVZdugkVwVWZE0GXRjcsK9gHtWDJppg3NZIYko9Rh/sONC0Xc+i/u/Bsz7Sy09UDgcM693JPjx4Jr4zjfqLSyura+Ud6sbG3v7O5V9w/aOkoUZS0aiUh1faKZ4JK1DDeCdWPFSOgL1vEn17nfeWBK80jemWnMvJCMJA84JcZK9/2QmLEfpE/ZIHWzQbXm1J0Z8DJxC1KDAs1B9as/jGgSMmmoIFr3XCc2XkqU4VSwrNJPNIsJnZAR61kqSci0l85SZ/jEKkMcRMo+afBM/b2RklDraejbyTylXvRy8T+vl5jg0ku5jBPDJJ0fChKBTYTzCvCQK0aNmFpCqOI2K6Zjogg1tqiKLcFd/PIyaZ/VXctvz2uNq6KOMhzBMZyCCxfQgBtoQgsoKHiGV3hDj+gFvaOP+WgJFTuH8Afo8wf5upLO</latexit><latexit sha1_base64="O4eplVAdWsWi2nj78Fp1SNWOEso=">AAAB9XicbVDLSgMxFL2pr1pfVZdugkVwVWZE0GXRjcsK9gHtWDJppg3NZIYko9Rh/sONC0Xc+i/u/Bsz7Sy09UDgcM693JPjx4Jr4zjfqLSyura+Ud6sbG3v7O5V9w/aOkoUZS0aiUh1faKZ4JK1DDeCdWPFSOgL1vEn17nfeWBK80jemWnMvJCMJA84JcZK9/2QmLEfpE/ZIHWzQbXm1J0Z8DJxC1KDAs1B9as/jGgSMmmoIFr3XCc2XkqU4VSwrNJPNIsJnZAR61kqSci0l85SZ/jEKkMcRMo+afBM/b2RklDraejbyTylXvRy8T+vl5jg0ku5jBPDJJ0fChKBTYTzCvCQK0aNmFpCqOI2K6Zjogg1tqiKLcFd/PIyaZ/VXctvz2uNq6KOMhzBMZyCCxfQgBtoQgsoKHiGV3hDj+gFvaOP+WgJFTuH8Afo8wf5upLO</latexit><latexit sha1_base64="O4eplVAdWsWi2nj78Fp1SNWOEso=">AAAB9XicbVDLSgMxFL2pr1pfVZdugkVwVWZE0GXRjcsK9gHtWDJppg3NZIYko9Rh/sONC0Xc+i/u/Bsz7Sy09UDgcM693JPjx4Jr4zjfqLSyura+Ud6sbG3v7O5V9w/aOkoUZS0aiUh1faKZ4JK1DDeCdWPFSOgL1vEn17nfeWBK80jemWnMvJCMJA84JcZK9/2QmLEfpE/ZIHWzQbXm1J0Z8DJxC1KDAs1B9as/jGgSMmmoIFr3XCc2XkqU4VSwrNJPNIsJnZAR61kqSci0l85SZ/jEKkMcRMo+afBM/b2RklDraejbyTylXvRy8T+vl5jg0ku5jBPDJJ0fChKBTYTzCvCQK0aNmFpCqOI2K6Zjogg1tqiKLcFd/PIyaZ/VXctvz2uNq6KOMhzBMZyCCxfQgBtoQgsoKHiGV3hDj+gFvaOP+WgJFTuH8Afo8wf5upLO</latexit><latexit sha1_base64="O4eplVAdWsWi2nj78Fp1SNWOEso=">AAAB9XicbVDLSgMxFL2pr1pfVZdugkVwVWZE0GXRjcsK9gHtWDJppg3NZIYko9Rh/sONC0Xc+i/u/Bsz7Sy09UDgcM693JPjx4Jr4zjfqLSyura+Ud6sbG3v7O5V9w/aOkoUZS0aiUh1faKZ4JK1DDeCdWPFSOgL1vEn17nfeWBK80jemWnMvJCMJA84JcZK9/2QmLEfpE/ZIHWzQbXm1J0Z8DJxC1KDAs1B9as/jGgSMmmoIFr3XCc2XkqU4VSwrNJPNIsJnZAR61kqSci0l85SZ/jEKkMcRMo+afBM/b2RklDraejbyTylXvRy8T+vl5jg0ku5jBPDJJ0fChKBTYTzCvCQK0aNmFpCqOI2K6Zjogg1tqiKLcFd/PIyaZ/VXctvz2uNq6KOMhzBMZyCCxfQgBtoQgsoKHiGV3hDj+gFvaOP+WgJFTuH8Afo8wf5upLO</latexit>

L1<latexit sha1_base64="oCcGK3z3uQSWxbsfTqHyOh3KcYo=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiGxcuKtgHtGPJpJk2NJMMSUYpw/yHGxeKuPVf3Pk3ZtpZaOuBwOGce7knJ4g508Z1v53Syura+kZ5s7K1vbO7V90/aGuZKEJbRHKpugHWlDNBW4YZTruxojgKOO0Ek+vc7zxSpZkU92YaUz/CI8FCRrCx0kM/wmYchOltNki9bFCtuXV3BrRMvILUoEBzUP3qDyVJIioM4VjrnufGxk+xMoxwmlX6iaYxJhM8oj1LBY6o9tNZ6gydWGWIQqnsEwbN1N8bKY60nkaBncxT6kUvF//zeokJL/2UiTgxVJD5oTDhyEiUV4CGTFFi+NQSTBSzWREZY4WJsUVVbAne4peXSfus7ll+d15rXBV1lOEIjuEUPLiABtxAE1pAQMEzvMKb8+S8OO/Ox3y05BQ7h/AHzucPsxySoA==</latexit><latexit sha1_base64="oCcGK3z3uQSWxbsfTqHyOh3KcYo=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiGxcuKtgHtGPJpJk2NJMMSUYpw/yHGxeKuPVf3Pk3ZtpZaOuBwOGce7knJ4g508Z1v53Syura+kZ5s7K1vbO7V90/aGuZKEJbRHKpugHWlDNBW4YZTruxojgKOO0Ek+vc7zxSpZkU92YaUz/CI8FCRrCx0kM/wmYchOltNki9bFCtuXV3BrRMvILUoEBzUP3qDyVJIioM4VjrnufGxk+xMoxwmlX6iaYxJhM8oj1LBY6o9tNZ6gydWGWIQqnsEwbN1N8bKY60nkaBncxT6kUvF//zeokJL/2UiTgxVJD5oTDhyEiUV4CGTFFi+NQSTBSzWREZY4WJsUVVbAne4peXSfus7ll+d15rXBV1lOEIjuEUPLiABtxAE1pAQMEzvMKb8+S8OO/Ox3y05BQ7h/AHzucPsxySoA==</latexit><latexit sha1_base64="oCcGK3z3uQSWxbsfTqHyOh3KcYo=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiGxcuKtgHtGPJpJk2NJMMSUYpw/yHGxeKuPVf3Pk3ZtpZaOuBwOGce7knJ4g508Z1v53Syura+kZ5s7K1vbO7V90/aGuZKEJbRHKpugHWlDNBW4YZTruxojgKOO0Ek+vc7zxSpZkU92YaUz/CI8FCRrCx0kM/wmYchOltNki9bFCtuXV3BrRMvILUoEBzUP3qDyVJIioM4VjrnufGxk+xMoxwmlX6iaYxJhM8oj1LBY6o9tNZ6gydWGWIQqnsEwbN1N8bKY60nkaBncxT6kUvF//zeokJL/2UiTgxVJD5oTDhyEiUV4CGTFFi+NQSTBSzWREZY4WJsUVVbAne4peXSfus7ll+d15rXBV1lOEIjuEUPLiABtxAE1pAQMEzvMKb8+S8OO/Ox3y05BQ7h/AHzucPsxySoA==</latexit><latexit sha1_base64="oCcGK3z3uQSWxbsfTqHyOh3KcYo=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiGxcuKtgHtGPJpJk2NJMMSUYpw/yHGxeKuPVf3Pk3ZtpZaOuBwOGce7knJ4g508Z1v53Syura+kZ5s7K1vbO7V90/aGuZKEJbRHKpugHWlDNBW4YZTruxojgKOO0Ek+vc7zxSpZkU92YaUz/CI8FCRrCx0kM/wmYchOltNki9bFCtuXV3BrRMvILUoEBzUP3qDyVJIioM4VjrnufGxk+xMoxwmlX6iaYxJhM8oj1LBY6o9tNZ6gydWGWIQqnsEwbN1N8bKY60nkaBncxT6kUvF//zeokJL/2UiTgxVJD5oTDhyEiUV4CGTFFi+NQSTBSzWREZY4WJsUVVbAne4peXSfus7ll+d15rXBV1lOEIjuEUPLiABtxAE1pAQMEzvMKb8+S8OO/Ox3y05BQ7h/AHzucPsxySoA==</latexit>

y1<latexit sha1_base64="Fgn0jFWpLK0PQHrwAbqZpb2JfnQ=">AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjRfsBbSib7aRdutmE3Y0QQn+CFw+KePUXefPfuG1z0NYXFh7emWFn3iARXBvX/XZKG5tb2zvl3cre/sHhUfX4pKPjVDFss1jEqhdQjYJLbBtuBPYShTQKBHaD6d283n1CpXksH02WoB/RseQhZ9RY6yEbesNqza27C5F18AqoQaHWsPo1GMUsjVAaJqjWfc9NjJ9TZTgTOKsMUo0JZVM6xr5FSSPUfr5YdUYurDMiYazsk4Ys3N8TOY20zqLAdkbUTPRqbW7+V+unJrz1cy6T1KBky4/CVBATk/ndZMQVMiMyC5QpbnclbEIVZcamU7EheKsnr0Pnqu5Zvr+uNZpFHGU4g3O4BA9uoAFNaEEbGIzhGV7hzRHOi/PufCxbS04xcwp/5Hz+AA8Gjac=</latexit><latexit sha1_base64="Fgn0jFWpLK0PQHrwAbqZpb2JfnQ=">AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjRfsBbSib7aRdutmE3Y0QQn+CFw+KePUXefPfuG1z0NYXFh7emWFn3iARXBvX/XZKG5tb2zvl3cre/sHhUfX4pKPjVDFss1jEqhdQjYJLbBtuBPYShTQKBHaD6d283n1CpXksH02WoB/RseQhZ9RY6yEbesNqza27C5F18AqoQaHWsPo1GMUsjVAaJqjWfc9NjJ9TZTgTOKsMUo0JZVM6xr5FSSPUfr5YdUYurDMiYazsk4Ys3N8TOY20zqLAdkbUTPRqbW7+V+unJrz1cy6T1KBky4/CVBATk/ndZMQVMiMyC5QpbnclbEIVZcamU7EheKsnr0Pnqu5Zvr+uNZpFHGU4g3O4BA9uoAFNaEEbGIzhGV7hzRHOi/PufCxbS04xcwp/5Hz+AA8Gjac=</latexit><latexit sha1_base64="Fgn0jFWpLK0PQHrwAbqZpb2JfnQ=">AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjRfsBbSib7aRdutmE3Y0QQn+CFw+KePUXefPfuG1z0NYXFh7emWFn3iARXBvX/XZKG5tb2zvl3cre/sHhUfX4pKPjVDFss1jEqhdQjYJLbBtuBPYShTQKBHaD6d283n1CpXksH02WoB/RseQhZ9RY6yEbesNqza27C5F18AqoQaHWsPo1GMUsjVAaJqjWfc9NjJ9TZTgTOKsMUo0JZVM6xr5FSSPUfr5YdUYurDMiYazsk4Ys3N8TOY20zqLAdkbUTPRqbW7+V+unJrz1cy6T1KBky4/CVBATk/ndZMQVMiMyC5QpbnclbEIVZcamU7EheKsnr0Pnqu5Zvr+uNZpFHGU4g3O4BA9uoAFNaEEbGIzhGV7hzRHOi/PufCxbS04xcwp/5Hz+AA8Gjac=</latexit><latexit sha1_base64="Fgn0jFWpLK0PQHrwAbqZpb2JfnQ=">AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjRfsBbSib7aRdutmE3Y0QQn+CFw+KePUXefPfuG1z0NYXFh7emWFn3iARXBvX/XZKG5tb2zvl3cre/sHhUfX4pKPjVDFss1jEqhdQjYJLbBtuBPYShTQKBHaD6d283n1CpXksH02WoB/RseQhZ9RY6yEbesNqza27C5F18AqoQaHWsPo1GMUsjVAaJqjWfc9NjJ9TZTgTOKsMUo0JZVM6xr5FSSPUfr5YdUYurDMiYazsk4Ys3N8TOY20zqLAdkbUTPRqbW7+V+unJrz1cy6T1KBky4/CVBATk/ndZMQVMiMyC5QpbnclbEIVZcamU7EheKsnr0Pnqu5Zvr+uNZpFHGU4g3O4BA9uoAFNaEEbGIzhGV7hzRHOi/PufCxbS04xcwp/5Hz+AA8Gjac=</latexit>

w1<latexit sha1_base64="CXTCUiruHEWUWGfg2nhR+EtfVYA=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9gFtKJPppB06mYSZG6WE/oYbF4q49Wfc+TdO2iy09cDA4Zx7uWdOkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeTaiFg94CzhfkTHSoSCUbTSYBBRnARh9jQfesNqza27C5B14hWkBgVaw+rXYBSzNOIKmaTG9D03QT+jGgWTfF4ZpIYnlE3pmPctVTTixs8WmefkwiojEsbaPoVkof7eyGhkzCwK7GSe0ax6ufif108xvPUzoZIUuWLLQ2EqCcYkL4CMhOYM5cwSyrSwWQmbUE0Z2poqtgRv9cvrpHNV9yy/v641mkUdZTiDc7gED26gAU1oQRsYJPAMr/DmpM6L8+58LEdLTrFzCn/gfP4AKqmRxQ==</latexit><latexit sha1_base64="CXTCUiruHEWUWGfg2nhR+EtfVYA=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9gFtKJPppB06mYSZG6WE/oYbF4q49Wfc+TdO2iy09cDA4Zx7uWdOkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeTaiFg94CzhfkTHSoSCUbTSYBBRnARh9jQfesNqza27C5B14hWkBgVaw+rXYBSzNOIKmaTG9D03QT+jGgWTfF4ZpIYnlE3pmPctVTTixs8WmefkwiojEsbaPoVkof7eyGhkzCwK7GSe0ax6ufif108xvPUzoZIUuWLLQ2EqCcYkL4CMhOYM5cwSyrSwWQmbUE0Z2poqtgRv9cvrpHNV9yy/v641mkUdZTiDc7gED26gAU1oQRsYJPAMr/DmpM6L8+58LEdLTrFzCn/gfP4AKqmRxQ==</latexit><latexit sha1_base64="CXTCUiruHEWUWGfg2nhR+EtfVYA=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9gFtKJPppB06mYSZG6WE/oYbF4q49Wfc+TdO2iy09cDA4Zx7uWdOkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeTaiFg94CzhfkTHSoSCUbTSYBBRnARh9jQfesNqza27C5B14hWkBgVaw+rXYBSzNOIKmaTG9D03QT+jGgWTfF4ZpIYnlE3pmPctVTTixs8WmefkwiojEsbaPoVkof7eyGhkzCwK7GSe0ax6ufif108xvPUzoZIUuWLLQ2EqCcYkL4CMhOYM5cwSyrSwWQmbUE0Z2poqtgRv9cvrpHNV9yy/v641mkUdZTiDc7gED26gAU1oQRsYJPAMr/DmpM6L8+58LEdLTrFzCn/gfP4AKqmRxQ==</latexit><latexit sha1_base64="CXTCUiruHEWUWGfg2nhR+EtfVYA=">AAAB83icbVDLSsNAFL2pr1pfVZduBovgqiQi6LLgpssK9gFtKJPppB06mYSZG6WE/oYbF4q49Wfc+TdO2iy09cDA4Zx7uWdOkEhh0HW/ndLG5tb2Tnm3srd/cHhUPT7pmDjVjLdZLGPdC6jhUijeRoGS9xLNaRRI3g2md7nffeTaiFg94CzhfkTHSoSCUbTSYBBRnARh9jQfesNqza27C5B14hWkBgVaw+rXYBSzNOIKmaTG9D03QT+jGgWTfF4ZpIYnlE3pmPctVTTixs8WmefkwiojEsbaPoVkof7eyGhkzCwK7GSe0ax6ufif108xvPUzoZIUuWLLQ2EqCcYkL4CMhOYM5cwSyrSwWQmbUE0Z2poqtgRv9cvrpHNV9yy/v641mkUdZTiDc7gED26gAU1oQRsYJPAMr/DmpM6L8+58LEdLTrFzCn/gfP4AKqmRxQ==</latexit>

m<latexit sha1_base64="qWck30ONVt2kTy0KbJVkwBgxduc=">AAAB6HicbZBNSwMxEIZn/az1q+rRS7AInsquCHoseOmxBfsB7VKy6Wwbm2SXJCuU0l/gxYMiXv1J3vw3pu0etPWFwMM7M2TmjVLBjfX9b29jc2t7Z7ewV9w/ODw6Lp2ctkySaYZNlohEdyJqUHCFTcutwE6qkcpIYDsa38/r7SfUhifqwU5SDCUdKh5zRq2zGrJfKvsVfyGyDkEOZchV75e+eoOEZRKVZYIa0w381IZTqi1nAmfFXmYwpWxMh9h1qKhEE04Xi87IpXMGJE60e8qShft7YkqlMRMZuU5J7cis1ubmf7VuZuO7cMpVmllUbPlRnAliEzK/mgy4RmbFxAFlmrtdCRtRTZl12RRdCMHqyevQuq4Ejhs35Wotj6MA53ABVxDALVShBnVoAgOEZ3iFN+/Re/HevY9l64aXz5zBH3mfP9ddjPc=</latexit><latexit sha1_base64="qWck30ONVt2kTy0KbJVkwBgxduc=">AAAB6HicbZBNSwMxEIZn/az1q+rRS7AInsquCHoseOmxBfsB7VKy6Wwbm2SXJCuU0l/gxYMiXv1J3vw3pu0etPWFwMM7M2TmjVLBjfX9b29jc2t7Z7ewV9w/ODw6Lp2ctkySaYZNlohEdyJqUHCFTcutwE6qkcpIYDsa38/r7SfUhifqwU5SDCUdKh5zRq2zGrJfKvsVfyGyDkEOZchV75e+eoOEZRKVZYIa0w381IZTqi1nAmfFXmYwpWxMh9h1qKhEE04Xi87IpXMGJE60e8qShft7YkqlMRMZuU5J7cis1ubmf7VuZuO7cMpVmllUbPlRnAliEzK/mgy4RmbFxAFlmrtdCRtRTZl12RRdCMHqyevQuq4Ejhs35Wotj6MA53ABVxDALVShBnVoAgOEZ3iFN+/Re/HevY9l64aXz5zBH3mfP9ddjPc=</latexit><latexit sha1_base64="qWck30ONVt2kTy0KbJVkwBgxduc=">AAAB6HicbZBNSwMxEIZn/az1q+rRS7AInsquCHoseOmxBfsB7VKy6Wwbm2SXJCuU0l/gxYMiXv1J3vw3pu0etPWFwMM7M2TmjVLBjfX9b29jc2t7Z7ewV9w/ODw6Lp2ctkySaYZNlohEdyJqUHCFTcutwE6qkcpIYDsa38/r7SfUhifqwU5SDCUdKh5zRq2zGrJfKvsVfyGyDkEOZchV75e+eoOEZRKVZYIa0w381IZTqi1nAmfFXmYwpWxMh9h1qKhEE04Xi87IpXMGJE60e8qShft7YkqlMRMZuU5J7cis1ubmf7VuZuO7cMpVmllUbPlRnAliEzK/mgy4RmbFxAFlmrtdCRtRTZl12RRdCMHqyevQuq4Ejhs35Wotj6MA53ABVxDALVShBnVoAgOEZ3iFN+/Re/HevY9l64aXz5zBH3mfP9ddjPc=</latexit><latexit sha1_base64="qWck30ONVt2kTy0KbJVkwBgxduc=">AAAB6HicbZBNSwMxEIZn/az1q+rRS7AInsquCHoseOmxBfsB7VKy6Wwbm2SXJCuU0l/gxYMiXv1J3vw3pu0etPWFwMM7M2TmjVLBjfX9b29jc2t7Z7ewV9w/ODw6Lp2ctkySaYZNlohEdyJqUHCFTcutwE6qkcpIYDsa38/r7SfUhifqwU5SDCUdKh5zRq2zGrJfKvsVfyGyDkEOZchV75e+eoOEZRKVZYIa0w381IZTqi1nAmfFXmYwpWxMh9h1qKhEE04Xi87IpXMGJE60e8qShft7YkqlMRMZuU5J7cis1ubmf7VuZuO7cMpVmllUbPlRnAliEzK/mgy4RmbFxAFlmrtdCRtRTZl12RRdCMHqyevQuq4Ejhs35Wotj6MA53ABVxDALVShBnVoAgOEZ3iFN+/Re/HevY9l64aXz5zBH3mfP9ddjPc=</latexit>

n<latexit sha1_base64="ZNZRLPnkv2vG9S2HtXEnuV36TT8=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4KXHFuwHtKFstpN27WYTdjdCCf0FXjwo4tWf5M1/47bNQVtfWHh4Z4adeYNEcG1c99spbG3v7O4V90sHh0fHJ+XTs46OU8WwzWIRq15ANQousW24EdhLFNIoENgNpveLevcJleaxfDCzBP2IjiUPOaPGWi05LFfcqrsU2QQvhwrkag7LX4NRzNIIpWGCat333MT4GVWGM4Hz0iDVmFA2pWPsW5Q0Qu1ny0Xn5Mo6IxLGyj5pyNL9PZHRSOtZFNjOiJqJXq8tzP9q/dSEd37GZZIalGz1UZgKYmKyuJqMuEJmxMwCZYrbXQmbUEWZsdmUbAje+smb0LmpepZbt5V6I4+jCBdwCdfgQQ3q0IAmtIEBwjO8wpvz6Lw4787HqrXg5DPn8EfO5w/Y4Yz4</latexit><latexit sha1_base64="ZNZRLPnkv2vG9S2HtXEnuV36TT8=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4KXHFuwHtKFstpN27WYTdjdCCf0FXjwo4tWf5M1/47bNQVtfWHh4Z4adeYNEcG1c99spbG3v7O4V90sHh0fHJ+XTs46OU8WwzWIRq15ANQousW24EdhLFNIoENgNpveLevcJleaxfDCzBP2IjiUPOaPGWi05LFfcqrsU2QQvhwrkag7LX4NRzNIIpWGCat333MT4GVWGM4Hz0iDVmFA2pWPsW5Q0Qu1ny0Xn5Mo6IxLGyj5pyNL9PZHRSOtZFNjOiJqJXq8tzP9q/dSEd37GZZIalGz1UZgKYmKyuJqMuEJmxMwCZYrbXQmbUEWZsdmUbAje+smb0LmpepZbt5V6I4+jCBdwCdfgQQ3q0IAmtIEBwjO8wpvz6Lw4787HqrXg5DPn8EfO5w/Y4Yz4</latexit><latexit sha1_base64="ZNZRLPnkv2vG9S2HtXEnuV36TT8=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4KXHFuwHtKFstpN27WYTdjdCCf0FXjwo4tWf5M1/47bNQVtfWHh4Z4adeYNEcG1c99spbG3v7O4V90sHh0fHJ+XTs46OU8WwzWIRq15ANQousW24EdhLFNIoENgNpveLevcJleaxfDCzBP2IjiUPOaPGWi05LFfcqrsU2QQvhwrkag7LX4NRzNIIpWGCat333MT4GVWGM4Hz0iDVmFA2pWPsW5Q0Qu1ny0Xn5Mo6IxLGyj5pyNL9PZHRSOtZFNjOiJqJXq8tzP9q/dSEd37GZZIalGz1UZgKYmKyuJqMuEJmxMwCZYrbXQmbUEWZsdmUbAje+smb0LmpepZbt5V6I4+jCBdwCdfgQQ3q0IAmtIEBwjO8wpvz6Lw4787HqrXg5DPn8EfO5w/Y4Yz4</latexit><latexit sha1_base64="ZNZRLPnkv2vG9S2HtXEnuV36TT8=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4KXHFuwHtKFstpN27WYTdjdCCf0FXjwo4tWf5M1/47bNQVtfWHh4Z4adeYNEcG1c99spbG3v7O4V90sHh0fHJ+XTs46OU8WwzWIRq15ANQousW24EdhLFNIoENgNpveLevcJleaxfDCzBP2IjiUPOaPGWi05LFfcqrsU2QQvhwrkag7LX4NRzNIIpWGCat333MT4GVWGM4Hz0iDVmFA2pWPsW5Q0Qu1ny0Xn5Mo6IxLGyj5pyNL9PZHRSOtZFNjOiJqJXq8tzP9q/dSEd37GZZIalGz1UZgKYmKyuJqMuEJmxMwCZYrbXQmbUEWZsdmUbAje+smb0LmpepZbt5V6I4+jCBdwCdfgQQ3q0IAmtIEBwjO8wpvz6Lw4787HqrXg5DPn8EfO5w/Y4Yz4</latexit>

m+ n<latexit sha1_base64="2dJ0uIcIoKiJe5CBum0XIq5GL0g=">AAAB6nicbZBNSwMxEIZn/az1q+rRS7AIglB2RdBjwUuPFe0HtEvJprNtaJJdkqxQSn+CFw+KePUXefPfmLZ70NYXAg/vzJCZN0oFN9b3v7219Y3Nre3CTnF3b//gsHR03DRJphk2WCIS3Y6oQcEVNiy3AtupRiojga1odDert55QG56oRztOMZR0oHjMGbXOepCXqlcq+xV/LrIKQQ5lyFXvlb66/YRlEpVlghrTCfzUhhOqLWcCp8VuZjClbEQH2HGoqEQTTuarTsm5c/okTrR7ypK5+3tiQqUxYxm5Tknt0CzXZuZ/tU5m49twwlWaWVRs8VGcCWITMrub9LlGZsXYAWWau10JG1JNmXXpFF0IwfLJq9C8qgSO76/L1VoeRwFO4QwuIIAbqEIN6tAABgN4hld484T34r17H4vWNS+fOYE/8j5/AAoujaQ=</latexit><latexit sha1_base64="2dJ0uIcIoKiJe5CBum0XIq5GL0g=">AAAB6nicbZBNSwMxEIZn/az1q+rRS7AIglB2RdBjwUuPFe0HtEvJprNtaJJdkqxQSn+CFw+KePUXefPfmLZ70NYXAg/vzJCZN0oFN9b3v7219Y3Nre3CTnF3b//gsHR03DRJphk2WCIS3Y6oQcEVNiy3AtupRiojga1odDert55QG56oRztOMZR0oHjMGbXOepCXqlcq+xV/LrIKQQ5lyFXvlb66/YRlEpVlghrTCfzUhhOqLWcCp8VuZjClbEQH2HGoqEQTTuarTsm5c/okTrR7ypK5+3tiQqUxYxm5Tknt0CzXZuZ/tU5m49twwlWaWVRs8VGcCWITMrub9LlGZsXYAWWau10JG1JNmXXpFF0IwfLJq9C8qgSO76/L1VoeRwFO4QwuIIAbqEIN6tAABgN4hld484T34r17H4vWNS+fOYE/8j5/AAoujaQ=</latexit><latexit sha1_base64="2dJ0uIcIoKiJe5CBum0XIq5GL0g=">AAAB6nicbZBNSwMxEIZn/az1q+rRS7AIglB2RdBjwUuPFe0HtEvJprNtaJJdkqxQSn+CFw+KePUXefPfmLZ70NYXAg/vzJCZN0oFN9b3v7219Y3Nre3CTnF3b//gsHR03DRJphk2WCIS3Y6oQcEVNiy3AtupRiojga1odDert55QG56oRztOMZR0oHjMGbXOepCXqlcq+xV/LrIKQQ5lyFXvlb66/YRlEpVlghrTCfzUhhOqLWcCp8VuZjClbEQH2HGoqEQTTuarTsm5c/okTrR7ypK5+3tiQqUxYxm5Tknt0CzXZuZ/tU5m49twwlWaWVRs8VGcCWITMrub9LlGZsXYAWWau10JG1JNmXXpFF0IwfLJq9C8qgSO76/L1VoeRwFO4QwuIIAbqEIN6tAABgN4hld484T34r17H4vWNS+fOYE/8j5/AAoujaQ=</latexit><latexit sha1_base64="2dJ0uIcIoKiJe5CBum0XIq5GL0g=">AAAB6nicbZBNSwMxEIZn/az1q+rRS7AIglB2RdBjwUuPFe0HtEvJprNtaJJdkqxQSn+CFw+KePUXefPfmLZ70NYXAg/vzJCZN0oFN9b3v7219Y3Nre3CTnF3b//gsHR03DRJphk2WCIS3Y6oQcEVNiy3AtupRiojga1odDert55QG56oRztOMZR0oHjMGbXOepCXqlcq+xV/LrIKQQ5lyFXvlb66/YRlEpVlghrTCfzUhhOqLWcCp8VuZjClbEQH2HGoqEQTTuarTsm5c/okTrR7ypK5+3tiQqUxYxm5Tknt0CzXZuZ/tU5m49twwlWaWVRs8VGcCWITMrub9LlGZsXYAWWau10JG1JNmXXpFF0IwfLJq9C8qgSO76/L1VoeRwFO4QwuIIAbqEIN6tAABgN4hld484T34r17H4vWNS+fOYE/8j5/AAoujaQ=</latexit>

><latexit sha1_base64="z2reKb9cCaLKO78JDhNmnvponyQ=">AAACCnicbVC7SgNBFL0bXzG+Vi1tRoNgIWFXBC1EAjYpLCKYByRLmJ3MJkNmH8zMCuuS2sZfsbFQxNYvsPNvnE22iIkHBs49517m3uNGnEllWT9GYWl5ZXWtuF7a2Nza3jF395oyjAWhDRLyULRdLClnAW0opjhtR4Ji3+W05Y5uMr/1QIVkYXCvkog6Ph4EzGMEKy31zMOrpMdOUdfHauh66eN4trrV1XXPLFsVawK0SOyclCFHvWd+d/shiX0aKMKxlB3bipSTYqEY4XRc6saSRpiM8IB2NA2wT6WTTk4Zo2Ot9JEXCv0ChSbq7ESKfSkT39Wd2ZJy3svE/7xOrLxLJ2VBFCsakOlHXsyRClGWC+ozQYniiSaYCKZ3RWSIBSZKp1fSIdjzJy+S5lnF1vzuvFyt5XEU4QCO4ARsuIAq1KAODSDwBC/wBu/Gs/FqfBif09aCkc/swx8YX7+Nfpow</latexit><latexit sha1_base64="z2reKb9cCaLKO78JDhNmnvponyQ=">AAACCnicbVC7SgNBFL0bXzG+Vi1tRoNgIWFXBC1EAjYpLCKYByRLmJ3MJkNmH8zMCuuS2sZfsbFQxNYvsPNvnE22iIkHBs49517m3uNGnEllWT9GYWl5ZXWtuF7a2Nza3jF395oyjAWhDRLyULRdLClnAW0opjhtR4Ji3+W05Y5uMr/1QIVkYXCvkog6Ph4EzGMEKy31zMOrpMdOUdfHauh66eN4trrV1XXPLFsVawK0SOyclCFHvWd+d/shiX0aKMKxlB3bipSTYqEY4XRc6saSRpiM8IB2NA2wT6WTTk4Zo2Ot9JEXCv0ChSbq7ESKfSkT39Wd2ZJy3svE/7xOrLxLJ2VBFCsakOlHXsyRClGWC+ozQYniiSaYCKZ3RWSIBSZKp1fSIdjzJy+S5lnF1vzuvFyt5XEU4QCO4ARsuIAq1KAODSDwBC/wBu/Gs/FqfBif09aCkc/swx8YX7+Nfpow</latexit><latexit sha1_base64="z2reKb9cCaLKO78JDhNmnvponyQ=">AAACCnicbVC7SgNBFL0bXzG+Vi1tRoNgIWFXBC1EAjYpLCKYByRLmJ3MJkNmH8zMCuuS2sZfsbFQxNYvsPNvnE22iIkHBs49517m3uNGnEllWT9GYWl5ZXWtuF7a2Nza3jF395oyjAWhDRLyULRdLClnAW0opjhtR4Ji3+W05Y5uMr/1QIVkYXCvkog6Ph4EzGMEKy31zMOrpMdOUdfHauh66eN4trrV1XXPLFsVawK0SOyclCFHvWd+d/shiX0aKMKxlB3bipSTYqEY4XRc6saSRpiM8IB2NA2wT6WTTk4Zo2Ot9JEXCv0ChSbq7ESKfSkT39Wd2ZJy3svE/7xOrLxLJ2VBFCsakOlHXsyRClGWC+ozQYniiSaYCKZ3RWSIBSZKp1fSIdjzJy+S5lnF1vzuvFyt5XEU4QCO4ARsuIAq1KAODSDwBC/wBu/Gs/FqfBif09aCkc/swx8YX7+Nfpow</latexit><latexit sha1_base64="z2reKb9cCaLKO78JDhNmnvponyQ=">AAACCnicbVC7SgNBFL0bXzG+Vi1tRoNgIWFXBC1EAjYpLCKYByRLmJ3MJkNmH8zMCuuS2sZfsbFQxNYvsPNvnE22iIkHBs49517m3uNGnEllWT9GYWl5ZXWtuF7a2Nza3jF395oyjAWhDRLyULRdLClnAW0opjhtR4Ji3+W05Y5uMr/1QIVkYXCvkog6Ph4EzGMEKy31zMOrpMdOUdfHauh66eN4trrV1XXPLFsVawK0SOyclCFHvWd+d/shiX0aKMKxlB3bipSTYqEY4XRc6saSRpiM8IB2NA2wT6WTTk4Zo2Ot9JEXCv0ChSbq7ESKfSkT39Wd2ZJy3svE/7xOrLxLJ2VBFCsakOlHXsyRClGWC+ozQYniiSaYCKZ3RWSIBSZKp1fSIdjzJy+S5lnF1vzuvFyt5XEU4QCO4ARsuIAq1KAODSDwBC/wBu/Gs/FqfBif09aCkc/swx8YX7+Nfpow</latexit>

F1<latexit sha1_base64="bjc15rgFOVXzCFeTU92prUoYcLw=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SL1sUK25dXcGtEy8gtSgQHNQ/eoPJUkiKgzhWOue58bGT7EyjHCaVfqJpjEmEzyiPUsFjqj201nqDJ1YZYhCqewTBs3U3xspjrSeRoGdzFPqRS8X//N6iQkv/ZSJODFUkPmhMOHISJRXgIZMUWL41BJMFLNZERljhYmxRVVsCd7il5dJ+6zuWX53XmtcFXWU4QiO4RQ8uIAG3EITWkBAwTO8wpvz5Lw4787HfLTkFDuH8AfO5w+p5pKa</latexit><latexit sha1_base64="bjc15rgFOVXzCFeTU92prUoYcLw=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SL1sUK25dXcGtEy8gtSgQHNQ/eoPJUkiKgzhWOue58bGT7EyjHCaVfqJpjEmEzyiPUsFjqj201nqDJ1YZYhCqewTBs3U3xspjrSeRoGdzFPqRS8X//N6iQkv/ZSJODFUkPmhMOHISJRXgIZMUWL41BJMFLNZERljhYmxRVVsCd7il5dJ+6zuWX53XmtcFXWU4QiO4RQ8uIAG3EITWkBAwTO8wpvz5Lw4787HfLTkFDuH8AfO5w+p5pKa</latexit><latexit sha1_base64="bjc15rgFOVXzCFeTU92prUoYcLw=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SL1sUK25dXcGtEy8gtSgQHNQ/eoPJUkiKgzhWOue58bGT7EyjHCaVfqJpjEmEzyiPUsFjqj201nqDJ1YZYhCqewTBs3U3xspjrSeRoGdzFPqRS8X//N6iQkv/ZSJODFUkPmhMOHISJRXgIZMUWL41BJMFLNZERljhYmxRVVsCd7il5dJ+6zuWX53XmtcFXWU4QiO4RQ8uIAG3EITWkBAwTO8wpvz5Lw4787HfLTkFDuH8AfO5w+p5pKa</latexit><latexit sha1_base64="bjc15rgFOVXzCFeTU92prUoYcLw=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SL1sUK25dXcGtEy8gtSgQHNQ/eoPJUkiKgzhWOue58bGT7EyjHCaVfqJpjEmEzyiPUsFjqj201nqDJ1YZYhCqewTBs3U3xspjrSeRoGdzFPqRS8X//N6iQkv/ZSJODFUkPmhMOHISJRXgIZMUWL41BJMFLNZERljhYmxRVVsCd7il5dJ+6zuWX53XmtcFXWU4QiO4RQ8uIAG3EITWkBAwTO8wpvz5Lw4787HfLTkFDuH8AfO5w+p5pKa</latexit>

F<latexit sha1_base64="u3UpL/haE0foGspep/c5hnxIc/k=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SFk2qNbcujsDWiZeQWpQoDmofvWHkiQRFYZwrHXPc2Pjp1gZRjjNKv1E0xiTCR7RnqUCR1T76Sx1hk6sMkShVPYJg2bq740UR1pPo8BO5in1opeL/3m9xISXfspEnBgqyPxQmHBkJMorQEOmKDF8agkmitmsiIyxwsTYoiq2BG/xy8ukfVb3LL87rzWuijrKcATHcAoeXEADbqEJLSCg4Ble4c15cl6cd+djPlpyip1D+APn8wf+/pLS</latexit><latexit sha1_base64="u3UpL/haE0foGspep/c5hnxIc/k=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SFk2qNbcujsDWiZeQWpQoDmofvWHkiQRFYZwrHXPc2Pjp1gZRjjNKv1E0xiTCR7RnqUCR1T76Sx1hk6sMkShVPYJg2bq740UR1pPo8BO5in1opeL/3m9xISXfspEnBgqyPxQmHBkJMorQEOmKDF8agkmitmsiIyxwsTYoiq2BG/xy8ukfVb3LL87rzWuijrKcATHcAoeXEADbqEJLSCg4Ble4c15cl6cd+djPlpyip1D+APn8wf+/pLS</latexit><latexit sha1_base64="u3UpL/haE0foGspep/c5hnxIc/k=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SFk2qNbcujsDWiZeQWpQoDmofvWHkiQRFYZwrHXPc2Pjp1gZRjjNKv1E0xiTCR7RnqUCR1T76Sx1hk6sMkShVPYJg2bq740UR1pPo8BO5in1opeL/3m9xISXfspEnBgqyPxQmHBkJMorQEOmKDF8agkmitmsiIyxwsTYoiq2BG/xy8ukfVb3LL87rzWuijrKcATHcAoeXEADbqEJLSCg4Ble4c15cl6cd+djPlpyip1D+APn8wf+/pLS</latexit><latexit sha1_base64="u3UpL/haE0foGspep/c5hnxIc/k=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuiIC4r2Ae0Y8mkmTY0kwxJRinD/IcbF4q49V/c+Tdm2llo64HA4Zx7uScniDnTxnW/ndLK6tr6RnmzsrW9s7tX3T9oa5koQltEcqm6AdaUM0FbhhlOu7GiOAo47QST69zvPFKlmRT3ZhpTP8IjwUJGsLHSQz/CZhyE6U02SFk2qNbcujsDWiZeQWpQoDmofvWHkiQRFYZwrHXPc2Pjp1gZRjjNKv1E0xiTCR7RnqUCR1T76Sx1hk6sMkShVPYJg2bq740UR1pPo8BO5in1opeL/3m9xISXfspEnBgqyPxQmHBkJMorQEOmKDF8agkmitmsiIyxwsTYoiq2BG/xy8ukfVb3LL87rzWuijrKcATHcAoeXEADbqEJLSCg4Ble4c15cl6cd+djPlpyip1D+APn8wf+/pLS</latexit>

+<latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit><latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit><latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit><latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit>

+<latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit><latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit><latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit><latexit sha1_base64="IwzizajS5AAFEtGqeN2O1uVSZxA=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tFEISSiKDHgpceW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219YeHhnRl25g0SwbVx3W+nsLG5tb1T3C3t7R8cHpWPT9o6ThXDFotFrLoB1Si4xJbhRmA3UUijQGAnmNzP650nVJrH8sFME/QjOpI85IwaazWvBuWKW3UXIuvg5VCBXI1B+as/jFkaoTRMUK17npsYP6PKcCZwVuqnGhPKJnSEPYuSRqj9bLHojFxYZ0jCWNknDVm4vycyGmk9jQLbGVEz1qu1uflfrZea8M7PuExSg5ItPwpTQUxM5leTIVfIjJhaoExxuythY6ooMzabkg3BWz15HdrXVc9y86ZSq+dxFOEMzuESPLiFGtShAS1ggPAMr/DmPDovzrvzsWwtOPnMKfyR8/kDc1WMtQ==</latexit>

H<latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit><latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit><latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit><latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit>

W<latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit><latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit><latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit><latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit>

H<latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit><latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit><latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit><latexit sha1_base64="OpZ/wlfQRvxCFObR9t4IFdACW+c=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEfRY8NJjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9GFR7z6h0jyWLTNL0I/oWPKQM2qs1awPyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhvZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXWyuMowgVcwjV4cAc1qEMD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8wei5Yze</latexit>

W<latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit><latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit><latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit><latexit sha1_base64="3Jxaq9KpHJuAf14Xe7x0WRj+bOI=">AAAB6HicbZBNS8NAEIYn9avWr6pHL4tF8FQSEeqx4MVjC/2CNpTNdtKu3WzC7kYoob/AiwdFvPqTvPlv3LY5aOsLCw/vzLAzb5AIro3rfjuFre2d3b3ifung8Oj4pHx61tFxqhi2WSxi1QuoRsEltg03AnuJQhoFArvB9H5R7z6h0jyWLTNL0I/oWPKQM2qs1ewOyxW36i5FNsHLoQK5GsPy12AUszRCaZigWvc9NzF+RpXhTOC8NEg1JpRN6Rj7FiWNUPvZctE5ubLOiISxsk8asnR/T2Q00noWBbYzomai12sL879aPzXhnZ9xmaQGJVt9FKaCmJgsriYjrpAZMbNAmeJ2V8ImVFFmbDYlG4K3fvImdG6qnuXmbaXeyuMowgVcwjV4UIM6PEAD2sAA4Rle4c15dF6cd+dj1Vpw8plz+CPn8we5oYzt</latexit>

Object Composer

Figure 4. Object feature map composition. The object category

is first encoded by a word embedding. Then the object feature

map is simply composed by filling the region within the object

bounding box with the concatenation of category embedding and

latent code. The rest of the feature map are all zeros. Symbol⊕stands for the vector concatenation, and

⊗means replicating

object representation within a bounding box.

integrate object feature maps one-by-one, and the last out-

put of the cLSTM is used as the fused hidden layout H,

which incorporates the location and category information

of all objects. Please refer to the supplementary material

for the structure of multi-layer cLSTM networks.

3.4. Image Decoder

Given the fused image hidden feature map H, image de-

coder is tasked with generating a result image. As shown

in Figure 2, there are two paths (blue and red) in the net-

works. They differ in latent code estimation. The blue path

reconstructs the input image using the object latent codes zrsampled from the posteriors Q(zr|O) that are conditioned

on the objects O in the input image I, while in the red one,

the latent codes zs are directly sampled from prior distri-

butions N (zs). As a result, two images are generated, i.e.,

I and I′, through the red and blue paths, respectively. Al-

through they may differ in appearance, both of them share

the same layout.

3.5. Object Latent Code Regression

To explicitly encourage the consistent connection be-

tween the latent codes and outputs, our model also tries to

recover the random sampled latent codes from the objects

generated along the red path. One can think of this as an

inference network for the latent codes. This helps prevent

a many-to-one mapping from the latent code to the output

during training, and as a result, produces more diverse re-

sults.

To achieve this, we use the same input object bounding

boxes L to crop the objects O′ in the generated image I

′.

The resized O′ are then sent to an object latent code es-

timator (which shares weights with the one used in image

reconstruction path), getting the estimated mean and vari-

ance vectors for the generated objects. We directly use the

8587

computed mean vectors, as the regressed latent codes z′

s,

and compare them with the sampled ones zs, for all objects.

3.6. Image and Object Discriminators

To make the generated images realistic, and the objects

recognizable, we adopt a pair of discriminators Dimg and

Dobj. The discriminator is trained to classify an input x or

y as real or fake by maximizing the objective [7]:

LGAN = Ex∼preal

logD(x) + Ey∼pfake

log(1−D(y)), (1)

where x represents the real images and y represents the gen-

erated ones. Meanwhile, the generator networks are trained

to minimizing LGAN. The image discriminator Dimg is ap-

plied to input images I, reconstructed images I and sampled

images I′, classifying them as real or fake.

The object discriminator Dobj is designed to assess the

quality and category of the real objects O, reconstructed

objects O and sampled objects O′ at the same time. In

addition, since O and O′ are cropped from the recon-

structed/sampled images according to the input bounding

boxes L, Dobj also encourages the generated objects to ap-

pear in their desired locations.

3.7. Loss Function

We end-to-end train the generator network and two dis-

criminator networks in an adversarial manner. The gener-

ator network, with all described components, is trained to

minimize the weighted sum of six losses:

• KL Loss LKL =∑o

i=1 E[DKL(Q(zri|Oi)||N (zr))]computes the KL-Divergence between the distribution

Q(zr|O) and the normal distribution N (zr), where o is

the number of objects in the image/layout.

• Image Reconstruction Loss Limg1 = ||I − I||1 penal-

izes the L1 difference between ground-truth image I and

reconstructed image I.

• Object Latent Code Reconstruction Loss Llatent1 =∑o

i=1 ||zsi − z′

si||1 penalizes the L1 difference between

the randomly sampled zs ∼ N(zs) and the re-estimated

z′

s from the generated objects O′.

• Image Adversarial Loss LimgGAN is defined as in Eq. (1),

where x is the ground truth image I, y is the reconstructed

image I and sampled image I′.

• Object Adversarial Loss LobjGAN is also defined as in

Eq. (1), where x is the objects O cropped from the ground

truth image I, y are O and O′ cropped from the recon-

structed image I and sampled image I′.

• Auxiliar Classification Loss LobjAC from Dobj encourages

the generated objects Oi and O′

i to be recognizable as

their corresponding categories.

Dataset Train Val. Test # Obj. # Obj. in Image

COCO [1] 24,972 1,024 2,048 171 3 ∼ 8

VG [19] 62,565 5,506 5,088 178 3 ∼ 30

Table 1. Statistics of COCO-Stuff and Visual Genome dataset.

Therefore, the final loss function of our model is defined as:

L =λ1LKL + λ2Limg1 + λ3L

latent1 +

λ4Limgadv + λ5L

objadv + λ6L

objAC,

where, λi are the parameters balancing different losses.

3.8. Implementation Details

We use SN-GAN [28] for stable training. Batch normal-

ization [12] and ReLU are used in the object encoder, image

decoder, and only ReLU is used in the discriminators (no

batch normalization). Conditional batch normalization [4]

is used in the object estimator to better normalize the object

feature map according to its category. After object fuser, we

use six residual blocks [8] to further refine the hidden image

feature maps. We set both m and n to 64. The image and

crop size are set to 64 × 64 and 32 × 32, respectively. The

λ1 ∼ λ6 are set to 0.01, 1, 10, 1, 1 and 1 respectively.

We train all models using Adam [17] with learning rate

of 0.0001 and batch size of 8 for 300,000 iterations; training

takes about 3 days on a single Titan Xp GPU. Full details

about our architecture can be found in the supplementary

material, and code will be made publicly available.

4. Experiments

Extensive experiments are conducted to evaluate the pro-

posed Layout2Im network. We first compare our proposed

method with previous state-of-the-art models for scene im-

age synthesis, and show its superiority in aspects of realism,

recognition and diversity. Finally, the contributions of each

loss for training our model are studied through ablation.

4.1. Datasets

The same as previous scene image generation

method [14], we evaluate our proposed model on the

COCO-Stuff [1] and Visual Genome [19] datasets. We

preprocess and split the two datasets the same as that

in [14]. Table 1 lists the datasets statistics. Each image

in these datasets has multiple bounding boxes annotations

with labels for the objects.

4.2. Baselines

We compare our approach with two state-of-the-art

methods: pix2pix [13] and sg2im [14].

pix2pix [13] translates images between two domains. In

this paper, we define the input domain as feature maps con-

structed from layout L, and set the real images as the output

domain. We construct the input feature map with the size of

8588

Layout

pix2pix

sg2im

Ours

Layout

pix2pix

sg2im

Ours

(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k)

(l) (m) (n) (o) (p) (q) (r) (s) (t) (u) (v)

boat

boat

sea

clouds buiding

skycraper

bridge

mountain

horseperson

tree

clothes

grass

dirt

tree

cowcow

grass

clouds cloudsairplane

tree

mountain

building

mirror

wall

sink

toilet

person

person

playfield fence personwall

foodplastic

person

person

dirt

person

elephantelephant

dirtbush

bush

giraffe

cloudssky

giraffe

handpizza

boy

hair

hand

door man hair

chair

floor

shirt

hand

cap

shirtroad

man

elephant

trunk

grass

building

buiding

tree

shadow

road

sky

ear

elephant

field hill

grass

grass

boy

hair

shadow

skateboard grass

clouds

water

track

grass

sky

line

sky

man

pant

mountain

ground

man

pant

tieshirt

head

tile

wall

mirror

sink

tilefloor

wall

treeclouds

hill

giraffe

grass

giraffewood

bushrock

zebra zebra

zebra

dirt

grass

skisnow

Figure 5. Examples of 64 × 64 generated images from complex layouts on COCO-Stuff (top) and Visual Genome Datasets (bottom) by

our proposed method and baselines. For each example, we show the input layout, images generated by pix2pix, sg2im and our method.

Please zoom in to see the category of each object. The ground truth images and more examples can be found in the supplementary material.

sky

snow

sky

snow

tree

person

sky

snow

tree

person

sky

tree

snow

fire hydrant fire hydrant

road

house house

car car

car

river

grass

tree

river

grass

tree

sheep

sheep

sheep

sheepLayout

Results

road

grass grass

river

tree

river

tree

sheep

fire hydrant

(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k)

Figure 6. Example of generated images by adding or moving bounding boxes based on previous layout. Three groups of images,

(a)-(c), (d)-(g) and (h)-(k), are shown. In (g) and (k), original bounding boxes are drawn in dash. Please zoom in to see the category of

each object.

C × H × W for each layout L, where C is the number of

object categories, H×W is the image size. A bounding box

Oi with label yi will set the corresponding region within c-

th channel (the channel for category yi) of the feature map

to 1 and others are all 0. The pix2pix model is learned to

translate the generated feature maps to real images.

sg2im [14] is originally trained to generate images from

scene graphs. However, it can also generate images from

layout, simply replacing the predicted layout with ground

truth layout. We list the Inception Score of sg2im using

ground truth layouts as reported in their paper, and generate

the results for other comparisons using their released model

8589

La

yo

ut

Sa

mp

le 1

Sa

mp

le 2

Sa

mp

le 3

(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k)

clouds

sky

hill

mountain

snow

sky

tree

fire hydrant

pavement

buildingtree

person

snow

sky

house

person

dirt

grass

treekite

person

tree

grass

sky

playingfield

person

personperson

person

tree fence

sheep

tree

sky

grass

building

clouds

personperson

grass

truck

grass

kite

kite

sky

tree

fence

person

wall-panel

buiding

playingfield

Figure 7. Examples of diverse images generated from same layouts. For each layout, we sample 3 images. The generated images have

different appearances, but sharing the same layout. Please zoom in to see the category of each object.

trained with ground truth layout. In other words, the input

and training data for our and sg2im models is identical.

4.3. Evaluation Metrics

Plausible images generated from layout should meet

three requirements: be realistic, recognizable and di-

verse. Therefore we choose four different metrics, Incep-

tion Score (IS) [35], Frechet Inception Distance (FID) [9],

Object Classification Accuracy (Accu.) and Diversity

Score (DS) [48].

Inception Score [35] is adopted to measure the quality, as

well as diversity, of generated images. In our paper, we use

the pre-trained VGG-net [39] as the base model to compute

the inception scores for our model and the baselines.

Frechet Inception Distance [9] uses 2nd order information

of the final layer of the inception model, and calculates the

similarity of generated images to real ones. Frechet Incep-

tion Distance is more robust to noise than Inception Score.

Classification Accuracy measures the ability to generate

recognizable objects, which is an important criteria for our

task. We first train a ResNet-101 model [8] to classify ob-

jects. This is done using the real objects cropped and resized

from ground truth images in the training set of each dataset.

We then compute and report the object classification accu-

racy for objects in the generated images.

Diversity Score computes the perceptual similarity be-

tween two images in deep feature space. Different from the

inception score which reflects the diversity across the entire

generated images, diversity score measures the difference of

a pair of images generated from the same input. We use the

LPIPS metric [48] for diversity score, and use AlexNet [20]

for feature extraction as suggested in the paper.

4.4. Qualitative results

Figure 5 shows generated images using our method, as

well as baselines. From these examples it is clear that our

method can generate complex images with multiple objects,

and even multiple instances of the same object type. For

example, Figure 5(a) shows two boats, (c) shows two cows,

(e) and (r) contain two people.

These examples also show that our method generates

images which respect the location constraints of the input

bounding boxes, and the generated objects in the image are

also recognizable and consistent with their input labels.

As we can see in Figure 5, pix2pix fails to generate

meaningful images, due to the extreme difficulty of directly

mapping layout to a real image without detailed instance

segmentation. The results generated by sg2im are also not

as good as ours. For example, in Figure 5 (g) and (i), the

generated giraffe and zebra are difficult to recognize, and

(l) contains lots of artifacts, making result look unrealistic.

In Figure 6 we demonstrate our model’s ability to gen-

erate complex images by starting with simple layout and

progressively adding new bounding box or moving exist-

ing bounding box, e.g., (g) and (k), to build/manipulate a

complex image. From these examples we can see that new

objects are drawn in the images at the desired locations, and

existing objects are kept consistent as new content is added.

Figure 7 shows the diverse results generated from the

same layouts. Given that the same layout may have many

different possible real image realizations, the ability to sam-

ple diverse images is a key advantage of our model.

4.5. Quantitative results

Table 2 summarizes comparison results of the inception

score, object classification accuracy and diversity score of

baseline models and our model. We also report the incep-

8590

IS FID Accu. DS

Method COCO VG COCO VG COCO VG COCO VG

Real Images (64 × 64) 16.3 ± 0.4 13.9 ± 0.5 - - 55.16 49.13 - -

pix2pix [13] 3.5 ± 0.1 2.7 ± 0.02 121.97 142.86 12.06 9.20 0 0

sg2im (GT Layout) [14] 7.3 ± 0.1 6.3 ± 0.2 67.96 74.61 30.04 40.29 0.02 ± 0.01 0.15 ± 0.12

Ours 9.1 ± 0.1 8.1 ± 0.1 38.14 31.25 50.84 48.09 0.15 ± 0.06 0.17 ± 0.09

Table 2. Performance on COCO and VG in Inception Score (IS), Frechet Inception Distance (FID), Object Classification Accu-

racy (Accu.) and Diversity Score (DS). The output size of all methods is 64 × 64. We train the pix2pix from scratch, and generate image

from the released sg2im model using ground truth layout.

Method IS Accu. DS

w/o Limg1 7.6 ± 0.2 49.03 0.17 ± 0.09

w/o Llatent1 7.5 ± 0.1 48.90 0.16 ± 0.09

w/o LobjAC 6.5 ± 0.1 10.06 0.37 ± 0.11

w/o Limgadv 7.1 ± 0.1 56.17 0.13 ± 0.09

w/o Lobjadv 7.3 ± 0.1 57.74 0.14 ± 0.09

full model 8.1 ± 0.1 48.09 0.17 ± 0.09

Table 3. Ablation study of our model on Visual Genome dataset

by removing different objectives. IS is the inception score, Accu.

is the object classification accuracy, and DS is the diversity score.

tion score and object classification accuracy on real images.

The proposed method significantly outperforms base-

lines in all the three evaluation metrics. In terms of In-

ception Score and Frechet Inception Distance, our method

outperforms the existing approaches with a substantial mar-

gin, presumably because our method generates more recog-

nizable objects as proved by object classification accuracy.

Please note that the object accuracy on real images is not

the upper bound of object classification accuracy, since the

object cannot be classified correctly in a real image does not

necessarily mean it is also difficult to distinguish in a gener-

ated image. Since the pix2pix is deterministic, its diversity

score is 0. By adding global noise to scene layout, sg2im

can generate images with limited diversity. The diversity

performance shows that our method can generate diverse

results from the same layout. A very notable improvement

is on COCO, where we achieve diversity score of 0.15 as

compared to 0.02 for sg2im.

4.6. Ablation Study

We demonstrate the necessity of all components of our

model by comparing the inception score, object classifica-

tion accuracy, and diversity score of several ablated versions

of our model trained on Visual Genome dataset:

• w/o Limg1 reconstructs ground truth images without pixel

regression.

• w/o Llatent1 does not regress the latent codes which are

used to generated objects in the result images.

• w/o LobjAC does not classify the category of objects.

• w/o Limgadv removes the object adversarial loss when train-

ing the model.

• w/o Lobjadv removes the image adversarial loss when train-

ing the model.

As shown in Table 3, removing any loss term will de-

crease the overall performance. Specifically, The model

trained without Limg1 or Llatent

1 generates less realistic im-

ages, which decreases the inception score. The object clas-

sification accuracy is still high because of the object clas-

sification loss. Without the constraint on reconstructed im-

ages or latent codes, the models get lower inception scores,

but similar diversity scores. Removing the object classifica-

tion loss degrade the inception score and object classifica-

tion accuracy significantly, since the model cannot generate

recognizable objects. Not surprisingly, this freedom results

in higher diversity score. It is expected to see that removing

the adversarial loss on image or object will decrease the in-

ception score substantially. However, the object classifica-

tion accuracy increases further comparing to the full model.

We believe that without the realism requirement of image

or object, the object classification loss could be tampered

with adversarial attack. Trained with all the losses, our full

model achieves a good balance across all three metrics.

5. Conclusion

In this paper we have introduced an end-to-end method

for generating diverse images from layout (bounding

boxes + categories). Our method can generate reasonable

images which look realistic and contain recognizable ob-

jects at the desired locations. We also showed that we can

control the image generation process by adding/moving ob-

jects in the layout easily. Qualitative and quantitative re-

sults on COCO-Stuff [1] and Visual Genome [19] datasets

demonstrated our model’s ability to generate realistic com-

plex images. Generating high resolution images from lay-

outs will be our future work. Moreover, making the image

generation process more controllable, such as specifying the

fine-grained attributes of instances, would be an interesting

future direction.

Acknowledgement This research was supported, in part, by

NSERC Discovery, NSERC DAS and NSERC CFI grants.

We gratefully acknowledge the support of NVIDIA Corpora-

tion with the donation of the Titan V GPU used for this re-

search.

8591

References

[1] Holger Caesar, Jasper Uijlings, and Vittorio Ferrari. Coco-

stuff: thing and stuff classes in context. arXiv: 1612.03716,

2016. 1, 2, 5, 8

[2] Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya

Sutskever, and Pieter Abbeel. Infogan: interpretable rep-

resentation learning by information maximizing generative

adversarial nets. In NIPS, 2016. 2

[3] Brian Cheung, Jesse A. Livezey, Arjun K. Bansal, and

Bruno A. Olshausen. Discovering hidden factors of varia-

tion in deep networks. In ICLR workshop, 2015. 2

[4] Harm de Vries, Florian Strub, Jeremie Mary, Hugo

Larochelle, Olivier Pietquin, and Aaron Courville. Modu-

lating early visual processing by language. In NIPS, 2017.

5

[5] Emily Denton and Vighnesh Birodkar. Unsupervised learn-

ing of disentangled representations from video. In NIPS,

2017. 2

[6] Alexey Dosovitskiy, Jost Tobias Springenberg, and Thomas

Brox. Learning to generate chairs with convolutional neural

networks. In CVPR, 2015. 2

[7] Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing

Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and

Yoshua Bengio. Generative adversarial networks. In NIPS,

2014. 5

[8] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun.

Deep residual learning for image recognition. In CVPR,

2016. 5, 7

[9] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner,

Bernhard Nessler, and Sepp Hochreiter. Gans trained by a

two time-scale update rule converge to a local nash equilib-

rium. In NIPS, 2017. 7

[10] Sepp Hochreiter and Jurgen Schmidhuber. Long short-term

memory. Neural Comput., 9(8):1735–1780, 1997. 4

[11] Seunghoon Hong, Dingdong Yang, Jongwook Choi, and

Honglak Lee. Inferring semantic layout for hierarchical text-

to-image synthesis. In CVPR, 2018. 1, 2

[12] Sergey Ioffe and Christian Szegedy. Batch normalization:

accelerating deep network training by reducing internal co-

variate shift. arXiv:1502.03167, 2015. 5

[13] Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A

Efros. Image-to-image translation with conditional adver-

sarial networks. In CVPR, 2017. 2, 5, 8

[14] Justin Johnson, Agrim Gupta, and Li Fei-Fei. Image genera-

tion from scene graphs. In CVPR, 2018. 1, 2, 5, 6, 8

[15] Levent Karacan, Zeynep Akata, Aykut Erdem, and Erkut Er-

dem. Learning to generate images of outdoor scenes from

attributes and semantic layouts. arXiv:1612.00215, 2016. 1,

2

[16] Jin-Hwa Kim, Devi Parikh, Dhruv Batra, Byoung-Tak

Zhang, and Yuandong Tian. Codraw: visual dialog for col-

laborative drawing. arXiv:1712.05558, 2017. 2

[17] Diederik P Kingma and Jimmy Ba. Adam: a method for

stochastic optimization. arXiv:1412.6980, 2014. 5

[18] Diederik P Kingma and Max Welling. Auto-encoding varia-

tional bayes. In ICLR, 2014. 2

[19] Ranjay Krishna, Yuke Zhu, Oliver Groth, Justin Johnson,

Kenji Hata, Joshua Kravitz, Stephanie Chen, Yannis Kalan-

tidis, Li-Jia Li, David A Shamma, et al. Visual genome:

connecting language and vision using crowdsourced dense

image annotations. IJCV, 2017. 1, 2, 5, 8

[20] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton.

Imagenet classification with deep convolutional neural net-

works. In NIPS, 2012. 7

[21] Wei-Sheng Lai, Jia-Bin Huang, Narendra Ahuja, and Ming-

Hsuan Yang. Deep laplacian pyramid networks for fast and

accurate super-resolution. In CVPR, 2017. 2

[22] Hsin-Ying Lee, Hung-Yu Tseng, Jia-Bin Huang, Maneesh

Singh, and Ming-Hsuan Yang. Diverse image-to-image

translation via disentangled representations. In ECCV, 2018.

2

[23] Ming-Yu Liu, Thomas Breuel, and Jan Kautz. Unsupervised

image-to-image translation networks. In NIPS, 2017. 2

[24] Liqian Ma, Qianru Sun, Stamatios Georgoulis, Luc Van

Gool, Bernt Schiele, and Mario Fritz. Disentangled person

image generation. In CVPR, 2018. 2

[25] Elman Mansimov, Emilio Parisotto, Jimmy Lei Ba, and Rus-

lan Salakhutdinov. Generating images from captions with

attention. arXiv:1511.02793, 2015. 1, 2

[26] Michael Mathieu, Junbo Zhao, Pablo Sprechmann, Aditya

Ramesh, and Yann LeCun. Disentangling factors of variation

in deep representations using adversarial training. In NIPS,

2016. 2

[27] Mehdi Mirza and Simon Osindero. Conditional generative

adversarial nets. arXiv:1411.1784, 2014. 2

[28] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and

Yuichi Yoshida. Spectral normalization for generative ad-

versarial networks. In ICLR, 2018. 5

[29] Zak Murez, Soheil Kolouri, David Kriegman, Ravi Ra-

mamoorthi, and Kyungnam Kim. Image to image translation

for domain adaptation. In CVPR, 2018. 2

[30] M. Nilsback and A. Zisserman. Automated flower classifica-

tion over a large number of classes. In Indian Conference on

Computer Vision, Graphics Image Processing, 2008. 1

[31] Aaron van den Oord, Nal Kalchbrenner, and Ko-

ray Kavukcuoglu. Pixel recurrent neural networks.

arXiv:1601.06759, 2016. 2

[32] Deepak Pathak, Philipp Krahenbuhl, Jeff Donahue, Trevor

Darrell, and Alexei A. Efros. Context encoders: feature

learning by inpainting. In CVPR, 2016. 2

[33] Scott Reed, Aaron van den Oord, Nal Kalchbrenner, Ser-

gio Gomez Colmenarejo, Ziyu Wang, Dan Belov, and Nando

de Freitas. Parallel multiscale autoregressive density estima-

tion. arXiv:1703.03664, 2017. 1, 2

[34] Scott E Reed, Zeynep Akata, Santosh Mohan, Samuel Tenka,

Bernt Schiele, and Honglak Lee. Learning what and where

to draw. In NIPS, 2016. 2

[35] Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki

Cheung, Alec Radford, and Xi Chen. Improved techniques

for training gans. In NIPS, 2016. 7

[36] Patsorn Sangkloy, Jingwan Lu, Chen Fang, Fisher Yu, and

James Hays. Scribbler: controlling deep image synthesis

with sketch and color. In CVPR, 2017. 2

8592

[37] Shikhar Sharma, Dendi Suhubdy, Vincent Michalski,

Samira Ebrahimi Kahou, and Yoshua Bengio. Chat-

painter: improving text to image generation using dialogue.

arXiv:1802.08216, 2018. 2

[38] Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung,

Wai kin Wong, and Wang chun Woo. Convolutional lstm

network: a machine learning approach for precipitation now-

casting. In NIPS, 2015. 4

[39] Karen Simonyan and Andrew Zisserman. Very deep

convolutional networks for large-scale image recognition.

arXiv:1409.1556, 2014. 7

[40] Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning

structured output representation using deep conditional gen-

erative models. In NIPS, 2015. 2

[41] Fuwen Tan, Song Feng, and Vicente Ordonez. Text2scene:

generating abstract scenes from textual descriptions.

arXiv:1809.01110, 2018. 1, 2

[42] Aaron van den Oord, Nal Kalchbrenner, Lasse Espeholt,

Oriol Vinyals, Alex Graves, et al. Conditional image gen-

eration with pixelcnn decoders. In NIPS, 2016. 2

[43] Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao,

Jan Kautz, and Bryan Catanzaro. High-resolution image

synthesis and semantic manipulation with conditional gans.

arXiv:1711.11585, 2017. 2

[44] P. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Be-

longie, and P. Perona. Caltech-UCSD birds 200. Technical

Report CNS-TR-2010-001, California Institute of Technol-

ogy, 2010. 1

[45] Wenqi Xian, Patsorn Sangkloy, Varun Agrawal, Amit Raj,

Jingwan Lu, Chen Fang, Fisher Yu, and James Hays. Tex-

turegan: controlling deep image synthesis with texture

patches. In CVPR, 2018. 2

[46] Chao Yang, Xin Lu, Zhe Lin, Eli Shechtman, Oliver Wang,

and Hao Li. High-resolution image inpainting using multi-

scale neural patch synthesis. In CVPR, 2017. 2

[47] Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiaolei

Huang, Xiaogang Wang, and Dimitris Metaxas. Stackgan:

text to photo-realistic image synthesis with stacked genera-

tive adversarial networks. In ICCV, 2017. 1, 2

[48] Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman,

and Oliver Wang. The unreasonable effectiveness of deep

features as a perceptual metric. In CVPR, 2018. 7

[49] Weiwei Zhang, Jian Sun, and Xiaoou Tang. Cat head detec-

tion - how to effectively exploit shape and texture features.

In ECCV, 2008. 1

[50] Bo Zhao, Bo Chang, Zequn Jie, and Leonid Sigal. Modular

generative adversarial networks. In ECCV, 2018. 2

[51] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A

Efros. Unpaired image-to-image translation using cycle-

consistent adversarial networks. In ICCV, 2017. 2

[52] Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Dar-

rell, Alexei A Efros, Oliver Wang, and Eli Shechtman. To-

ward multimodal image-to-image translation. In NIPS, 2017.

2

8593