+ All Categories
Home > Technology > Transfer Learning Neural Network implementation for food recognition

Transfer Learning Neural Network implementation for food recognition

Date post: 20-Mar-2017
Category:
Upload: ali-abul-hawa
View: 76 times
Download: 2 times
Share this document with a friend
8
Estimating Calorie Content in Food Dishes using Transfer Learning Neural Networks Ali Abul Hawa
Transcript
Page 1: Transfer Learning Neural Network implementation for food recognition

Estimating Calorie Content in Food Dishes using Transfer Learning

Neural Networks

Ali Abul Hawa

Page 2: Transfer Learning Neural Network implementation for food recognition

2

Page 3: Transfer Learning Neural Network implementation for food recognition

Difficulty in this project/dataset: many food dishes look very similar

Beef noodleRamen noodle

3

Page 4: Transfer Learning Neural Network implementation for food recognition

4

Low-levelfeatures

Higher-levelfeatures

Output Classification

Layers

1.2M images1000 classes

16 Layers

VGG16, 16 layers: 3 weeks on 4 GPUs Titan Black ~$6000

My computer (~ g2.2xlarge AWS) : ~ 30 weeks

TRANSFER LEARNING

Page 5: Transfer Learning Neural Network implementation for food recognition

5

Low-levelfeatures

Higher-levelfeatures

CustomOutput

Classification Layers

15 Layers

100Food

classifier

Train custom classification layer ~52%

TRANSFER LEARNING

Page 6: Transfer Learning Neural Network implementation for food recognition

6

Low-levelfeatures

Higher-levelfeatures

CustomOutput

Classification Layers

15 Layers

100Food

classifier

Train classification layer + highest level feature layer

~ 70% top-1, 92% top-5 accuracyin 30 hours 30 weeks

TRANSFER LEARNING

Page 7: Transfer Learning Neural Network implementation for food recognition

7

TRANSFER LEARNING

Why transfer learning?Time

Works well even on small datasets. However, need to use image augmentations to prevent overfitting!

Page 8: Transfer Learning Neural Network implementation for food recognition

8

Live Demo …

… Thanks!


Recommended