Mashiro

Will centertrack work if the test video is long?

3

I found that centertrack use the pre_hm from the ground truth during training process. However, centertrack predicts the bbox and id using pre_hm calculated by previous frame. Actually, this pre_hm...

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...

[WIP] Support launch ddp by train.py

1

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...

[Fix] Enhance the compatibility of training stylegan 2

1

## Motivation PyTorch consolidated convolution backward API in [PR](https://github.com/pytorch/pytorch/commit/3dc3651e0ee3623f669c3a2c096408dbc476d122). It means users cannot train styegan2 with PyTorch version higher than 1.10.1 This PR enhances the compatibility of the code referring...

@biubug6 Thanks for your excellent work on the PyTorch implementation of RetinaFace, and I believe this project has helped many other developers study and apply RetinaFace :smile: . With the...

[FEATURE]: Is there any plan to support pure bf16 training for `GeminiDDP`?

2

### Describe the feature [Llama-2](https://github.com/facebookresearch/llama) has made `fsdp` + `bf16` training as their default training setting. The memory occupied by the copy of fp32 optimizer state and fp32 model parameters...

MMEngine community collaboration :rocket: !

23

# MMEngine community collaboration Interested in deeply participating in OpenMMLab projects? Want to learn more about OpenMMLab's awesome tools? MMEngine calls for volunteers from the community to participate in daily...

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...

[Bug] `accumulative_counts` does not apply to `parameter_schedulers`

4

The accurate cumulative pipeline should be as follows: ```python loss = model(data) loss.backward() if idx % accumulative_counts == 0: optimizer.step() scheduler.step() optimizer.zero_grad() ``` However, the condition `idx % accumulative_counts` is...

[New Config] Follow-up & Known issues

1

## Limitations The unsupported syntax for **New Config**, and these syntaxes will not be supported in the future version 1. The configuration file name must comply with the naming convention...