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

Can Biformer run in the current MMDet? And can this model utilize any of the detection heads in MMdet? #35

Closed MaiZiPiaoPiao closed 9 months ago

MaiZiPiaoPiao commented 9 months ago

Can Biformer run in the current MMDet? And can this model utilize any of the detection heads in MMdet?

rayleizhu commented 9 months ago

See https://github.com/rayleizhu/BiFormer/tree/public_release/object_detection

MaiZiPiaoPiao commented 9 months ago

谢谢您的解答，我看了您的代码，发现您是基于mmdet写出来的代码，对吧？ 我的问题是，能否直接将您所写的关于biformer的代码放入mmdet当中直接运行，而不需要进行修改？

rayleizhu commented 9 months ago

谢谢您的解答，我看了您的代码，发现您是基于mmdet写出来的代码，对吧？ 我的问题是，能否直接将您所写的关于biformer的代码放入mmdet当中直接运行，而不需要进行修改？

我不理解什么叫做”直接“，”不需要进行修改“。你似乎对mmdet有什么误解。实际上，按照我的理解，任何pytorch模型都可以直接放入mmdet运行，需要做的仅仅是

(1) 保证接口一致(i.e. forward输出一个list, 包含4个level的feature map) https://github.com/rayleizhu/BiFormer/blob/1697bbbeafb8680524898f1dcaac10defd0604be/object_detection/models_mm/biformer_mm.py#L56

(2) 加入mmdet的registry https://github.com/rayleizhu/BiFormer/blob/1697bbbeafb8680524898f1dcaac10defd0604be/object_detection/models_mm/biformer_mm.py#L11

可能特别一点的是，我的代码里使用了继承来消除冗余代码，突出对mmdet做的适配，懂python的话应该容易理解的： https://github.com/rayleizhu/BiFormer/blob/1697bbbeafb8680524898f1dcaac10defd0604be/object_detection/models_mm/biformer_mm.py#L12