WGAN with feedback from discriminator& LayerNorm instead of BatchNorm
- Added feedback from the discriminator in the form of attention 2.Modified BN to LayerNorm as BatchNorm creates correlation between the examples. Note that the results from LayerNorm were not promising
python main.py --dataset lsun --dataroot /workspace/lsun --cuda --clamp --fdback --save_dir sample
Need the lsun data. Follow the instruction in original implementation link
WGAN Model based on original paper link.
Cross Attention is my own implementation
Rerun and recheck the results