你好，可以用keras或者tensorflow实现一下CBN吗？还有就是直接用python实现的CBN速度上会不会收到影响呢？

Excuse me. I don't know the mean of 'weight' in forward function. I just want to use CBN in my work. Can anyone help me ?

The work is really inspiring, in terms of dealing with high variance of small mini-batch size BN statistics, while keeps the conventional property of BN fusion during inference.

However, I am confused with the following fraction of codes:

Assume self.buffer_num, self.burnin are given.
self.pre_mu was initialized as an empty list, and appended inside the "if condition scope" itself:

It seems that one would never reach the condition from scratch, for which len(self.pre_mu)==self.buffer_num doesn't hold.
Is there anything misunderstood? Thanks.

Hi, the question is in the title. THank you.

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/amFNsyUBvm or add me on WeChat (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.

It looks very similar to the effect of GN. I wonder if there is an advantage in forward speed?

I was using a resent18 model from torchvision，and try to replace all bn by CBN. But I can't train it at all, the accuracy is always very little, such as 0.01%

I also use an pre trained resnet18 model that has been trained for 40 epochs, and continue train with CBN replaced for some epochs, and the accuracy is also dropped immediately.

Is there any details I had missed？Please give me some help, Thanks.

Thanks for your excellent work but there's something in your 4.2 experiment that confused me.
Does the 'fixed BN' mean just keep the backbone's BN weights(beta&gamma) fixed or freeze all the backbone's weights?

thank you.