Update README.md
Browse files
README.md
CHANGED
|
@@ -13,7 +13,7 @@ Try out this model [here](https://huggingface.co/spaces/PrakhAI/AIPlane).
|
|
| 13 |
|  |  |
|
| 14 |
|
| 15 |
# Training Progression
|
| 16 |
-
<video width="
|
| 17 |
<source src="https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/qFlnTITZwS3DSTxLp0Oa8.mp4" type="video/mp4">
|
| 18 |
</video>
|
| 19 |
|
|
@@ -32,7 +32,7 @@ Spectral Normalization is a technique suggested for training GANs in [this paper
|
|
| 32 |
|
| 33 |
It aims to make the Critic's (Discriminator's) outputs mathematically continuous w.r.t. the space of input images, avoiding exploding gradients.
|
| 34 |
|
| 35 |
-
|
| 36 |
|
| 37 |
| Batch Normalization | Spectral Normalization |
|
| 38 |
| ----------- | ------------ |
|
|
@@ -46,4 +46,8 @@ For 32x32 images of Airplanes, even a short initial round of Progressive Growing
|
|
| 46 |
|
| 47 |
| Flat Growing | Progressive Growing |
|
| 48 |
| ----------- | ------------ |
|
| 49 |
-
|  |  |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|  |  |
|
| 14 |
|
| 15 |
# Training Progression
|
| 16 |
+
<video width="25%" controls>
|
| 17 |
<source src="https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/qFlnTITZwS3DSTxLp0Oa8.mp4" type="video/mp4">
|
| 18 |
</video>
|
| 19 |
|
|
|
|
| 32 |
|
| 33 |
It aims to make the Critic's (Discriminator's) outputs mathematically continuous w.r.t. the space of input images, avoiding exploding gradients.
|
| 34 |
|
| 35 |
+
Spectral Normalization works very well in practice to stabilize the training of the GAN, as demonstrated by the example below (comparison at equivalent points during training):
|
| 36 |
|
| 37 |
| Batch Normalization | Spectral Normalization |
|
| 38 |
| ----------- | ------------ |
|
|
|
|
| 46 |
|
| 47 |
| Flat Growing | Progressive Growing |
|
| 48 |
| ----------- | ------------ |
|
| 49 |
+
|  |  |
|
| 50 |
+
|
| 51 |
+
The generator for this model generates 4x4, 8x8, 16x16 and 32x32 images, which form the inputs for the critic. Each resolution is associated with a 'weight' (α<sub>4</sub>, α<sub>8</sub>, α<sub>16</sub>, α<sub>32</sub>), which indicate the focus on the corresponding image resolution at any given time during the training.
|
| 52 |
+
|
| 53 |
+
At the beginning of the training, α<sub>4</sub>=1, α<sub>8</sub>=0, α<sub>16</sub>=0, α<sub>32</sub>=0, with α<sub>4</sub>=0, α<sub>8</sub>=0, α<sub>16</sub>=0, α<sub>32</sub>=1 towards the end.
|