Update README.md
Browse files
README.md
CHANGED
|
@@ -7,8 +7,38 @@ datasets:
|
|
| 7 |
| ----- | --------- |
|
| 8 |
|  |  |
|
| 9 |
|
|
|
|
| 10 |
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
<video controls
|
|
|
|
| 7 |
| ----- | --------- |
|
| 8 |
|  |  |
|
| 9 |
|
| 10 |
+
This GAN model is trained on the [FGVC Aircraft](https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/) dataset. The model uses [Progressive Growing](https://arxiv.org/pdf/1710.10196.pdf) with [Spectral Normalization](https://arxiv.org/pdf/1802.05957.pdf).
|
| 11 |
|
| 12 |
+
The work builds up on https://huggingface.co/PrakhAI/AIPlane and https://huggingface.co/PrakhAI/AIPlane2.
|
| 13 |
+
|
| 14 |
+
This model was trained to generate 256x256 images of Aircrafts. The implementation in JAX on Colab can be found [here](https://colab.research.google.com/github/prakharbanga/AIPlane3/blob/main/AIPlane3_ProGAN_%2B_Spectral_Norm_(256x256).ipynb).
|
| 15 |
+
|
| 16 |
+
# Convolutional Architecture
|
| 17 |
+
|
| 18 |
+
A significant improvement over https://huggingface.co/PrakhAI/AIPlane2 is the elimination of "checkerboard" artifacts. This is done by using Image Resize followed by Convolution layer in the Generator instead of a Transposed Convolution where the kernel size is not divisible by the stride.
|
| 19 |
+
|
| 20 |
+
| Transposed Convolution (kernel size not divisible by stride) | Resize followed by convolution |
|
| 21 |
+
| - | - |
|
| 22 |
+
|  |  |
|
| 23 |
+
|
| 24 |
+
# Image Quality
|
| 25 |
+
|
| 26 |
+
The model, while generating several high quality images of Airplanes, also generates poor quality images.
|
| 27 |
|
| 28 |
+
A total of 400 generated images were labeled by hand as either desirable (151) or undesirable (249).
|
| 29 |
+
|
| 30 |
+
| Sample desirable outputs | Sample undesirable outputs |
|
| 31 |
+
| --------- | ------------ |
|
| 32 |
+
|  |  |
|
| 33 |
+
|
| 34 |
+
# Latent Space Interpolation
|
| 35 |
+
|
| 36 |
+
Latent Space Interpolation can an educational exercise to get deeper insight into the model.
|
| 37 |
+
|
| 38 |
+
It can be observed below that several aspects of the generated image such as the color of the sky, grounded-ness of the plane, as well as the plane shape and color are frequently continuous through the latent space.
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
# Training Progression
|
| 43 |
|
| 44 |
+
<video controls width="50%" src="https://cdn-uploads.huggingface.co/production/uploads/649f9483d76ca0fe679011c2/o2NDDMQPhdEY5Vc96b31G.mp4"></video>
|