search

NVlabs / PWC-Net

PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume, CVPR 2018 (Oral)
Other
1.63k stars 357 forks source link

PWC-Net-small #111

Open JerryLeolfl opened 4 years ago

JerryLeolfl commented 4 years ago

Hi Deqing, I am interested to your work PWC-net, and I want to use it for optical flow estimation.But I have something confused.

  1. I wonder that is that PWC-net-small in the paper means cut all skip connections in every dense block. I mean that if I want to use PWC-net-small, I will change code in PWC-Net/PyTorch/models/PWCNet.py /: x = torch.cat((corr2, c12, up_flow3, up_feat3), 1) x = torch.cat((self.conv2_0(x), x),1) x = torch.cat((self.conv2_1(x), x),1) x = torch.cat((self.conv2_2(x), x),1) x = torch.cat((self.conv2_3(x), x),1) x = torch.cat((self.conv2_4(x), x),1) to x = torch.cat((corr2, c12, up_flow3, up_feat3), 1) x = self.conv2_0(x) x = self.conv2_1(x) x = self.conv2_2(x) x = self.conv2_3(x) x = self.conv2_4(x) ...the same changes as what in the 3rd, 4th, 5th, 6th dense block Is that what you did for pwc-net-small?
  1. what is different between 'pwc_dc_net' and 'pwc_dc_net_old'?

I am looking forward to your responce. Thank you