Welcome update to OpenMMLab 2.0

I am Vansin, the technical operator of OpenMMLab. In September of last year, we announced the release of OpenMMLab 2.0 at the World Artificial Intelligence Conference in Shanghai. We invite you to upgrade your algorithm library to OpenMMLab 2.0 using MMEngine, which can be used for both research and commercial purposes. If you have any questions, please feel free to join us on the OpenMMLab Discord at https://discord.gg/A9dCpjHPfE or add me on WeChat (ID: van-sin) and I will invite you to the OpenMMLab WeChat group.

Here are the OpenMMLab 2.0 repos branches:

Attention: please create a new virtual environment for OpenMMLab 2.0.

Why do we clamp here?

Thank you for a nice and handy implementation. I would like to ask you to provide some kind of stylegan2 training config, e.g. like this one
so it would be possible to replicate your experiment. Most of all Im interested in understanding used combination of losses, it is not completely clear to me if you used ONLY focal frequency loss and not other losses in stylegan2 experiment. so would be cool to know relative weights of losses used.
thanks.

Could you please provide the method of reproducing other methods in the paper? This repo Only provide vanilla AE .

As I couldn't find a tensorflow implementation of Focal Frequency Loss, so I created it.

Please visit the Github Repo and PyPi Project.

Use case notebook is included in the Repo. Any feedback is appreciated.

@EndlessSora If you find this implementation useful, kindly do mention it on your README.

Thanks for releasing this.

Thanks for your good job!
recently i am apply it for my work, i find its well for Image Reconstruction. but i am confuse for how big its value ? Generally speaking，for gan, we will have two loss. can you provide some experience for two loss? should i initialize two loss is equal?

Hi, thanks for your work.
I got a problem when I use the focal frequency loss for training.
This sentence appears above the log file(but the network is still on training process):
Warning: Casting complex values to real discards the imaginary part (function operator())

Great work! Thank you for sharing the code.

Could you please share the code to draw the spectra of images? Thanks.

Can you tell me the reason and how can I mod codes to support pytorch>1.7.1 ?
For example 1.8.1 ?

Hello, will the gradient of loss be calculated when ffl is used as loss? I try to put ffl on cuda, but it is always a scalar, and mse is a tensor with calculated gradient. Here are my changes:


The following are the values of mse and ffl, their forms are different, I don’t know if I’m wrong here.

Could you please share the paper of Deceive D: Adaptive Pseudo Augmentation for GAN Training with Limited Data, THX.

The warning code is【C:\Users\PC.conda\envs\paGAN\lib\site-packages\torch\autograd_init_.py:173: UserWarning: Casting complex values to real discards the imaginary part (Triggered internally at C:\cb\pytorch_1000000000000\work\aten\src\ATen\native\Copy.cpp:239.)
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass】

Does it affect the quality of the generated images? Thank you!

Hi, thank you for providing the code. It was really helpful.

One thing I am curious about is that when training VAE, unlike VanillaAE, the KL loss weight can affect the recontruction quality.
Adding focal frequency loss without changing the weight for KL loss will casue the recontruction loss to be weighted more, so that it is trivially able to reconstruct more details.

Can you share about how did you weight these terms? I did not find this description in the paper.

Thanks in advance!