search

Visual-Attention-Network / VAN-Segmentation

Apache License 2.0
191 stars 31 forks source link

Invitation of making PR for MMSegmentation. #4

Open MengzhangLI opened 2 years ago

MengzhangLI commented 2 years ago

Hi, I am a member of OpenMMLab and mainly work for developing MMSegmentation.

This work is very great and if it supported officially, many more people would use it for benchmark, which would promote research in this area.

Would you like to make PR for openmmlab? i.e., refactoring some codes to keep the same style for our current code.

We would like to support you by using our own GPU to train.

I think it is pretty cool because it would make more reseachers and community members use this excellent work!

Looking forward to your reply!

Best,

MenghaoGuo commented 2 years ago

Hi,

Thanks for your attention to VAN.

We are willing to add models into mmsegmentation. I think we can communicate conveniently by WeChat.

Best, Menghao

MengzhangLI commented 2 years ago

OK, I have added you. Looking for your acceptance. ;)

I think we could first support VAN backbone into MMClassification, then downstream MMDetection and MMSegmentation could utilize this backbone. Here is related Chinese article in Zhihu.

Here are:

(1) Reference about ConvNext. In ConvNext, we refactored its original code and made pr in mmclassification. I think VAN could follow this procedure, i.e., you just make a pr in MMClassfication, and we would review it to refactor it into our usual code style.

Here are our current ConvNext in mmclassification and downstream MMSegmentation PR, which would be merged in this week.

(2) Making a pull request. Usefule links: Contributing to OpenMMLab and Chinese article from zhihu.

For Visual-Attention-Network, please refer to PR about supporting ConvNext in MMClassification. Refactoring is a little bit laborious but that would make users more convenient after this treatment. We would review it quickly to respond your warm-hearted help for us!

We need to first ensure current models (such as in ImageNet and ADE20K) could get same results in this PR, aka, aligning inference metric.

After that, we would train many models (for example, VAN + UperNet) to re-implement paper results. That would be implemented on our computation resources.

If you have any problems about usage or pr of MMSegmentation and MMClassification, feel free to contact us. Let's work together to provide VAN as a benchmark for community!

Best,

MenghaoGuo commented 2 years ago

Thanks. we will learn and try to PR as soon as possible.

shinya7y commented 2 years ago

Related PR in MMClassification. https://github.com/open-mmlab/mmclassification/pull/739