search

amanchadha / iSeeBetter

iSeeBetter: Spatio-Temporal Video Super Resolution using Recurrent-Generative Back-Projection Networks | Python3 | PyTorch | GANs | CNNs | ResNets | RNNs | Published in Springer Journal of Computational Visual Media, September 2020, Tsinghua University Press
https://arxiv.org/abs/2006.11161
MIT License
363 stars 68 forks source link

Output of APIT pretrained model has a mosaic pattern #11

Closed buildist closed 4 years ago

buildist commented 4 years ago

When using netG_epoch_4_1_APITLoss.pth the output is less smooth than the L1 loss model as expected, but there's a faint mosaic pattern all over everything. Did you notice this in your research or is there a problem with my setup?

image

amanchadha commented 4 years ago

Hey,

I'd love to help you figure this out, but I'll need more detail to figure out what's up. Can you send me the parameters you're testing with?

buildist commented 4 years ago

Thanks! I didn't train my own model using APIT loss option yet - simply used yours that is in weights/netG_epoch_4_1_APITLoss.pth. Just wondering if that's what you get when testing with that model.

amanchadha commented 4 years ago

Overall, your output looks reasonable to me. The pattern might also be an artifact of your test workload, but I think it looks in-line with the results I remember seeing when I was experimenting with iSeeBetter.

buildist commented 4 years ago

Cool, I'll do some more training experiments and see what I can figure out. Trying to do super-resolution combined with compression artifact reduction and I'm already getting good results from training my own model.

amanchadha commented 4 years ago

Good to hear! I'm open to collaborate so if you're interested in exploring that/something together, I'm down!