Zheng Hui (惠政)
Zheng Hui (惠政)
The bicubic interpolation algorithm in python opencv is different from that in matlab. I found PIL.image.resize() function can obtain approximate result. @akafen
@NJU-Jet Hello, it doesn't need to use L2 loss. I recommend that training more iterations on larger learning rates, like 2e-4 and 1e-4. The performance improvement is mainly in these...
@nan-rock https://github.com/Zheng222/IMDN/blob/f1170cfb51f86f25f68bafa80fce04b2470f48ed/test_IMDN_AS.py#L30-L60
@WangZhiFeng1996 Please refer to #15 .
@fantasysponge The previous pytorch (e.g., pytorch 1.1) did not have this warning. You can modify the location of adjusting learning rate by yourself.
@yuqiangchu You need train x2 model firstly and then use the pretrained ckpt to initialize the x4 model.
@yuqiangchu You can train the x2 model serveral times when the learning rate is 2e-4 or 1e-4.
@leiwen83 Of course, please refer to https://github.com/Zheng222/VideoHDR
@bhaveshIITB Hello, for irregular masks, the random regular regions are cropped and sent to the local discriminator.
@muxgt Of course, testing and training codes will be released successively. Thanks for your interest.