Inception-vae
Variational Auto Encoder using Inception module in PyTorch
Summary
Improving the blurring peculiar to VAE using the Inception module.
Inception VAE(Proposed method)
CNN VAE
Changing an inception module
Original inception is concatenation, but use simple addition instead.
Experiment
Name | Inception | # repeat | # params(M) | F/B pass size(MB) | batch_size |
---|---|---|---|---|---|
Normal-1 | No | 1 | 1.96 M | 227.07 | 32 |
Normal-2 | No | 2 | 3.53 M | 337.32 | 32 |
Normal-3 | No | 3 | 5.10 M | 447.57 | 32 |
Normal-4 | No | 4 | 6.67 M | 557.82 | 32 |
Normal-5 | No | 5 | 8.24 M | 668.07 | 32 |
Normal-6 | No | 6 | 9.81 M | 778.32 | 32 |
Normal-8 | No | 8 | 12.95 M | 998.82 | 16 |
Normal-10 | No | 10 | 16.09 M | 1219.32 | 16 |
Inception-1 | Yes | 1 | 5.95 M | 709.41 | 32 |
Inception-2 | Yes | 2 | 11.51 M | 1302.00 | 16 |
- Batch_size of Normal-08, Normal-10, and Inception-02 are halved due to GPU memory.
- F/B pass size and # params are calculated by pytorch-summary.
Result
See details(Japanese)
https://qiita.com/koshian2/items/e2d05d9151f5ae9deefb
Reference
This code was experimented with The Japanese Female Facial Expression (JAFFE) Database. Thank you.
Michael J. Lyons, Shigeru Akemastu, Miyuki Kamachi, Jiro Gyoba.
Coding Facial Expressions with Gabor Wavelets, 3rd IEEE International Conference on Automatic Face and Gesture Recognition, pp. 200-205 (1998).
Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich.
Going Deeper with Convolutions, CVPR2015