Post on 25-Jan-2017
transcript
Disclaimer
Please be advised that this presentation may contain content which some viewers may find offensive.
Disclaimer
Please be advised that this presentation may contain content which some viewers may find offensive.
convnet
Demo network for this talk (modified ZF net):
5 convolutional layers3 fully-connected layers2 softmax outputs (‘sfw’ and ‘nsfw’)
Deconvolutional Networks
• Provides way to map activations at high layers back to the input
• Same operations as Convnet, but in reverse:– Unpool feature maps– Convolve unpooled maps
• Filters copied from Convnet
– Used here purely as a probe– Originally proposed as unsupervised learning
method– No inference, no learning Input Image
Convolution (learned)
Unpooling
Feature maps
Non-linearity
[Zeiler et al. CVPR’10, ICCV’11]
Reversible Max Pooling
Pooled Feature Maps
MaxLocations“Switches”
Pooling Unpooling
Feature MapReconstructed Feature Map
First Layer
Image from Zeiler, Matthew D., and Rob Fergus. "Visualizing and understanding convolutional networks." European Conference on Computer Vision. Springer International Publishing, 2014.
https://en.wikipedia.org/wiki/Lenna
Lenna or Lena is the name given to a standard test image widely used in the field of image processing since 1973.[1] It is a picture of Lena Söderberg, shot by photographer Dwight Hooker, cropped from the centerfold of the November 1972 issue of Playboy magazine.
Data and Research
Data Science● Collect data● Improve
Models
Research● Solve problems
we don’t have solutions for today
What about data?
Datasets● 100+ datasets● Almost every
vertical where there are images and videos
Data● 1B+ labelled
images● 1M+ labelled
videos