icandle / efdn Goto Github PK

Hello,

MSU Graphics & Media Lab Video Group has recently launched two new Super-Resolution Benchmarks.

• Video Upscalers Benchmark: Quality Enhancement determines the best upscaling methods for increasing video resolution and improving visual quality.
• Super-Resolution for Video Compression benchmark aims to test Super-Resolution methods on compressed videos and select the best model for each video codec standard.

If you are interested in participating, you can add your algorithm following the submission steps:

• Submit for Video Upscalers Benchmark: Quality Enhancement
• Submit for Super-Resolution for Video Compression benchmark

We would be grateful for your feedback on our work!

can you tell me about you loss function code

hello, i can not find the place to set patch size, it is not involved in the train.yml?

EDBB depth_multiplier preset = out_planes * inp_planes > mid_planes = out_planes * depth_multiplierare right?
ECBSR depth_multiplier preset = 2.0

https://github.com/xindongzhang/ECBSR/blob/main/models/ecbsr.py

您好！您的这篇加强边缘特征的图像超分文章对我很有启发，但是我在阅读的过程中有几点疑问，想向您请教一下。
1.损失函数的权重确定

请问这里scaled parameters Sx,Sy,Sl是指什么？是EDBB中对应soblex, sobley, laplace3个分支conv1的卷积核参数吗？
还有normalized weights of s又是指什么？以及这个normalized weights是怎么计算的？
2. 梯度方差计算公式

按照我以往对方差计算的理解，我认为这里分子上应该有平方，请问您这里没有进行平方运算是如何考虑的？

感谢您在百忙之中阅读我的问题，期待您的回复！谢谢！

What parameter do I need to change for 2x upscaling?

Thanks for your work! I've learned a lot and read your core code. Now I have some questions in training details.
In the paper, during the training phase we use EDBB with 7 branches. When training finished we reparameterize EDBB into vanilla convolution for inference and deployment. These can be implemented by 'switch_to_deploy' method in 'EFDN' class.
What confused me most is:

1. what's the purpose of ‘switch_to_gv’ method in 'EFDN' class? As it discards conv1x1 and conv1x1_3x3 branches.
2. whether we need to use ‘switch_to_gv' during training phase?
   Looking forward to your reply!