search

GengDavid / pytorch-cpn

A PyTorch re-implementation of CPN (Cascaded Pyramid Network for Multi-Person Pose Estimation)
GNU General Public License v3.0
484 stars 98 forks source link

Refine Net #4

Closed Tiamo666 closed 6 years ago

Tiamo666 commented 6 years ago

I guess maybe there is some problems in implementation of refine net. In your refineNet.py, you define the forward pass as follows: def forward(self, x): refine_fms = [] for i in range(4): refine_fms.append(self.cascadei) out = torch.cat(refine_fms, dim=1) out = self.final_predict(out)
return out I think you should inverse the x, eg: x = x[::-1], because x[0] is the smallest feature map, and x[3] is biggest feature map. And there are 3 bottlenecks after smallest feature map , 0 bottleneck after biggest feature map according to paper.

GengDavid commented 6 years ago

Hi, @Tiamo666 Yes, you are right. It is suppose to have 3 bottlenecks after the smallest feature map and no bottleneck after the biggest feature map. Thanks for pointing it out! I'll correct it.

Tiamo666 commented 6 years ago

Thanks a lot for your reimplementation work.

GengDavid commented 6 years ago

I’ve updated the codes and results, so I am going to close this issue. If you still have any questions, feel free to reopen it.