Elvis Yu-Jing Lin

Sorry, I didn't check the issues for a long time. What I did in this repo was migrating the neural network from the TensorFlow version. The TensorFlow implementation does not...

Hi @morkap, which PyTorch version are you using? I only tested on 0.4.0 and everything should work well. If you train AttGAN with newer PyTorch version, I cannot guarantee the...

Briefly speaking, shortcut layers pass the information from encoder layers to decoder layers. Since the AttGAN generator, which is composed of an encoder and a decoder, is a U-Net structure,...

Not at all. However, I'm sorry that I removed all the stuffs of AttGAN from my local machine since I graduated. You'll have to re-train the model from scratch to...

Can you access [this google drive file](https://drive.google.com/drive/folders/1_E5YCb4XOTZpt6KBwBzSaJdofoqPViN8)? The file is 154MB. After you download and extract the zip, you will find it under `256_shortcut1_inject1_none_hq/checkpoint/weights.199.pth`.

The model is saved every 1000 iterations, so there are a bunch of model in the end. To reduce the disk storage they take, I only save the weights of...

Could you tell me what command you used to train it?

That was an dimension mismatch error in the AttGAN generator. When decoding the latent space vector, the conditional attributes were not concateanted to the latent space vector due to their...

That's interesting... Would you mind giving it a try with other PyTorch version such as 1.5.0? That's the version I tested AttGAN on CelebA dataset a few weeks ago.

Hi @EoinKenny, the problem seems not related to shortcut layers. I tried training with 6 layers of `enc_layers` and `dec_layers` an d it worked. But it crashed when I changed...