Results 5 comments of MrL-CV

i meet the same problem, but someone said it may be the mismatch between the cuda version, but i dont known how to solve it

> It seems `sigmoid + norm` is used by default. Adding `MODEL.SPARSE_INST.DECODER.NAME GroupIAMSoftDecoder` to the command line solved the problem for me. It still does not work...

> Hi @kirillkoncha, sorry for the late reply. Maybe you can fix it by: > > ```python > iam_prob = iam_prob.view(B, N, -1) > normalizer = iam_prob.sum(-1).float().clamp(min=1e-4) > iam_prob =...

> 楼主你好,请问此问题解决了吗? 我也是遇到这个问题,是在使用CLRNet-onnxruntime-and-tensorrt-demo这个项目将clr_head.py替换后重新训练出现的报错。 换回来就好了:)

> Pls check if the generated boundary is correct do you mean check the generated boundary on ground truth label files?