Yang Yang

def pytorch_normalze(img): """ https://github.com/pytorch/vision/issues/223 return appr -1~1 RGB """ tensor = t.from_numpy(img) dtype = tensor.dtype mean = [0.485, 0.456, 0.406] std = [0.229, 0.224, 0.225] mean = t.as_tensor(mean, dtype=dtype, device=tensor.device)...


I wrote the poink-linking-network based on your code,but the loss looked strange which leaped after every epoch.So I had took a close look out at your loss,I found that there...

As a two-stage model, Faster RCNN can train RPN and ROI individually using roi_cls_locs、roi_scores、rpn_locs, rpn_scores. But in your code，you only use the roi_cls_locs and roi_scores to train the ROI head，and...

I meet the same problem, could you tell me the way you solved the problem? Thanks.

@GYC1996 I met the same problem as you. Could you please tell me how you solved it later？

@zyg11 Have you reproduced the training results of the author？

https://github.com/chenjun2hao/CenterFace.pytorch/blob/37a351be07606136561bc56eae716f8575b61e00/src/lib/trains/multi_pose.py#L34 change to this: if not opt.mse_loss: output['hm'] = _sigmoid(output['hm'])

我想问一下这样的结果正常么，精度上的差距主要是因为缺失第三步骤么？还有这里的border可以带来多少的精度提升呢 ：） https://github.com/chenjun2hao/CenterFace.pytorch/blob/4ea7b08255f282d8ddbbf6c6011114018de07c74/src/lib/datasets/sample/multi_pose_2.py#L48

https://github.com/chenjun2hao/CenterFace.pytorch/blob/37a351be07606136561bc56eae716f8575b61e00/src/lib/datasets/sample/multi_pose.py#L15 去年的实验，我不太记得了。建议把训练的bbox值打出来看一下，此外可以看一下这个函数。