huiyegit

Thank you for your reply. The error has been solved, but the results are different from the ones in the paper. Is there anything wrong? 

Maybe. My experiments for the pretrain_DAMSM.py were conducted under the AttnGAN+CL.

Yes. Thank you for pointing out it. I have changed it.

There are the following things that you may have a try: 1. Use the correct source file to calculate FID score([link](https://github.com/MinfengZhu/DM-GAN/tree/master/eval/FID)); 2. Check the fake images and real images for...

To calculate the FID score, the command should be like this: **python fid_score.py --gpu 0 --batch-size 50 --path1 real_images_bird --path2 fake_images_bird** path1 is the directory containing the real images for...

It should be the epoch 200. You may also try other checkpoints.

The metric for DAMSM may be the IS score, FID for the image quality.

> What do you mean try other checkpoints ? > > What about `valid loss`, `s_loss`, `w_loss` outputs when I train the DAMSM, because after epoch 200, these stats seems...

You are right. The feedback of DAMSM model is after the GAN training.

For pretrained Inception-V3 on Bird, try this one: https://github.com/hanzhanggit/StackGAN-inception-model