Output images are completely black

[MFarooqAit]

Dear author,
Thank you so much for sharing the simplified code of ESRGAN.

I have used your training code for my own dataset, but testing code produced completely black super resolved images.

Kindly guide.
Best,
Farooq

[purijs]

Same issue, were you able to resolve it?

[MFarooqAit]

Dear purijis,
Try to load your trained model by editing the test.py file, e.g,
net.load_state_dict(torch.load('checkpoints/generator_100.pth'))

Results are still not very impressive.

[purijs]

Dear purijis,
Try to load your trained model by editing the test.py file, e.g,
net.load_state_dict(torch.load('checkpoints/generator_100.pth'))

Results are still not very impressive.

Thanks, also the loss values keep fluctuating for me, is it happening for you too? I tried training on 100 images for 500 epochs

[MFarooqAit]

I did not observed it, but results are not very good on my dataset (1600 images) trained for 200 epochs.

[purijs]

I did not observed it, but results are not very good on my dataset (1600 images) trained for 200 epochs.

Right, for me, model is also not generalising well for test data, what other approaches or gan models are you trying?

[thunanguyen]

I think we need to train around 100k epochs to have a good GAN guys

[thunanguyen]

Sr, I've edited to make it more clear Vào Th 4, 4 thg 11, 2020 vào lúc 16:37 Jaskaran Singh Puri < notifications@github.com> đã viết:

…

I think we need to train around 100k to have a good GAN guys 100K of what, epochs or images? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHZZ4WDENOAGQYM7Y5DHLYLSOEOE7ANCNFSM4PS5XGIQ> .

[veb-101]

Any updates on this?