xuanzhangyang
xuanzhangyang
> Has the mish cuda function been installed following the instruction: https://github.com/WongKinYiu/ScaledYOLOv4/blob/yolov4-large/README.md#installation? Yes, I can successfully inference yolov4-p6 model with https://github.com/WongKinYiu/ScaledYOLOv4/blob/yolov4-large engine.
> for pretrained models, these lines will work: > https://github.com/linghu8812/tensorrt_inference/blob/887cca1487395cc46a23537213201d224600a976/ScaledYOLOv4/export_onnx.py#L28-L41 The problem is solved. \# modify import in models/common.py \# line 7 -> from mc.build.lib.mish_mish import MishCuda as Mish The...
> 可以用embed替代断点。 > > ```python > from IPython import embed > embed() > exit() > ``` 我想通过单步调试来了解整个代码流程，想要使用embed做到这一点会非常麻烦。 pdb无法使用的主要是原因是使用了torch.distributed.launch， google了很多，没找到解决方法。
@zhiqi-li
> 知乎上有一位热心网友整理了代码的流程图，你可以参考。 bevformer代码解读 - Vidar的文章 - 知乎 https://zhuanlan.zhihu.com/p/529828000 @zhiqi-li 感谢志琦大佬，我先看看。
@zhiqi-li 大佬啥时候有空自己出一个代码流程梳理相关的博客，感觉用了mmdet3d框架后，代码难懂很多。
> @Robertwyq numworks设成0这个我清楚，然后还需要自己修改代码把分布式训练的部分去掉？
So I need to pick out the broken part of the picture， but sometimes it's hard to do
Hi @liuguilin1225 , do you plan to open source code?