search

rayleizhu / BiFormer

[CVPR 2023] Official code release of our paper "BiFormer: Vision Transformer with Bi-Level Routing Attention"
https://arxiv.org/abs/2303.08810
MIT License
460 stars 36 forks source link

Objection detect code #2

Closed C-C-Y closed 1 year ago

C-C-Y commented 1 year ago

Hello, may I ask when the code for object detection will be released? I look forward to your response. Thank you.

rayleizhu commented 1 year ago

The MM project series share the same backbone usage. Hence it is not difficult to transfer the segmentation code to detection.

As you asked, I will arrange the release in two weeks.

C-C-Y commented 1 year ago

The MM project series share the same backbone usage. Hence it is not difficult to transfer the segmentation code to detection.

As you asked, I will arrange the release in two weeks.

Thank you very much.

LMMMEng commented 1 year ago

Thank you for your wonderful work! Could you please release the config file for the detection first?

rayleizhu commented 1 year ago

Thank you for your wonderful work! Could you please release the config file for the detection first?

Maybe by next Friday. I cannot guarantee as there are many other things on my hand. But I'm sure the full detection code can be released in two weeks.

Hiyao-yy commented 1 year ago

Thanks for sharing, looking forward to the full detection code！

rayleizhu commented 1 year ago

The object detection code has been released.

Note that you need to update the environment if you have done setup previously.

LMMMEng commented 1 year ago

Thanks for your effort, I wonder if norm eval can improve performance https://github.com/rayleizhu/BiFormer/blob/8c6f0f7d908e9842d6dd6d11ed87507948d934fc/object_detection/models_mm/biformer_mm.py#L30 Moreover, could you let me know what the bugs are on the normalization layer?

rayleizhu commented 1 year ago

Thanks for your effort, I wonder if norm eval can improve performance

I did not remember how much it helps. You may have a try. If you have a conclusion, please tell me here. By the way, norm_eval=True is the default choice of mmdetetction:

https://github.com/rayleizhu/BiFormer/blob/8c6f0f7d908e9842d6dd6d11ed87507948d934fc/object_detection/configs/_base_/models/mask_rcnn_r50_fpn.py#L12

Moreover, could you let me know what the bugs are on the normalization layer?

See my comments here:

https://github.com/rayleizhu/BiFormer/blob/8c6f0f7d908e9842d6dd6d11ed87507948d934fc/object_detection/models_mm/biformer_mm.py#L50

Actually, I do not know if it is this modification that works. There are some other differences (not in my model but functions or environment provided by third parties), as I mentioned in the README.

LMMMEng commented 1 year ago

Thanks for your effort, I wonder if norm eval can improve performance

I did not remember how much it helps. You may have a try. If you have a conclusion, please tell me here. By the way, norm_eval=True is the default choice of mmdetetction:

https://github.com/rayleizhu/BiFormer/blob/8c6f0f7d908e9842d6dd6d11ed87507948d934fc/object_detection/configs/_base_/models/mask_rcnn_r50_fpn.py#L12

Moreover, could you let me know what the bugs are on the normalization layer?

See my comments here:

https://github.com/rayleizhu/BiFormer/blob/8c6f0f7d908e9842d6dd6d11ed87507948d934fc/object_detection/models_mm/biformer_mm.py#L50

Actually, I do not know if it is this modification that works. There are some other differences (not in my model but functions or environment provided by third parties), as I mentioned in the README.

Many thanks!