-
Notifications
You must be signed in to change notification settings - Fork 25
Open
Description
Thanks for your repo.
I used the default training code of publayout, and trained for 10 epoches.
The final training loss reaches around 1.27.
epoch 10 iter 5245: train loss 1.27828. lr 2.880000e-04: 100%|██████████| 5246/5246 [46:58<00:00, 1.86it/s]
I opened the annotated code groups (total 4), and found that the input image is fine while the output seems not satisfactory.
for i, layout in enumerate(layouts): layout = self.test_dataset.render(layout) layout.save(os.path.join(self.config.samples_dir, f'recon_{epoch:02d}_{i:02d}.png'))
Any suggestions or comments will be appreciated. thanks!
Metadata
Metadata
Assignees
Labels
No labels
