I've checked out your class conditional generation (imagenet) and unconditional synthesis (celeba 512) and the results seem very promising. However, I am having trouble evaluating the downstream tasks.
Could you update / give clear instructions on how to run the inpainting downstream task? (for example, on CelebA-HQ 256?).
The args given in the readme don't all seem to match with the code in downstream_tasks/test_flow_latent_inpainting.py (for instance, the argparse accepts epoch_id, not num_epoch).
It also seems that there is a separate checkpoint used for CelebA Inpainting (i.e. model_500.pth) that is not given.
Hi guys,
I've checked out your class conditional generation (imagenet) and unconditional synthesis (celeba 512) and the results seem very promising. However, I am having trouble evaluating the downstream tasks.
Could you update / give clear instructions on how to run the inpainting downstream task? (for example, on CelebA-HQ 256?).
The args given in the readme don't all seem to match with the code in
downstream_tasks/test_flow_latent_inpainting.py
(for instance, the argparse acceptsepoch_id
, notnum_epoch
).It also seems that there is a separate checkpoint used for CelebA Inpainting (i.e. model_500.pth) that is not given.
Thanks!