PrakhAI commited on
Commit
9347d94
·
1 Parent(s): 834600d

Fix Progressive Growing justification

Browse files
Files changed (1) hide show
  1. README.md +3 -3
README.md CHANGED
@@ -30,12 +30,12 @@ The Generator uses Transposed Convolutions (2x2 kernel) with strides for upsampl
30
 
31
  Progressive Growing of GAN resolutions is suggested to improve the Quality and Stability of GAN training, especially for higher resolution models (1024x1024).
32
 
33
- For 32x32 images of Airplanes, it is unclear if there is a positive impact on the Image Quality at comparable points during model training:
34
 
35
  | Flat Growing (50K steps) | Progressive Growing (50K steps) |
36
  | ----------- | ------------ |
37
- | ![image.png](https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/CbqOqA3z08lWPoygrbk-w.png) | ![image.png](https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/lT_IwqkL60M1RSxEtJLaJ.png) |
38
 
39
- However, it doesn't negatively impact the training process and the additional parameter cost is very small (876.6 KB vs 855.1 KB for the generator).
40
 
41
  # Does the model simply memorize the images?
 
30
 
31
  Progressive Growing of GAN resolutions is suggested to improve the Quality and Stability of GAN training, especially for higher resolution models (1024x1024).
32
 
33
+ For 32x32 images of Airplanes, even a short initial round of Progressive Growing clearly has an impact:
34
 
35
  | Flat Growing (50K steps) | Progressive Growing (50K steps) |
36
  | ----------- | ------------ |
37
+ | ![image.png](https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/QnTET-5ae_0x11CcXeWgR.png) | ![image.png](https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/lT_IwqkL60M1RSxEtJLaJ.png) |
38
 
39
+ Moreover, the additional parameter cost is very small (876.6 KB vs 855.1 KB for the generator).
40
 
41
  # Does the model simply memorize the images?