alexhex7 / deformconv_pytorch

Deformable Convolutional Networks

Deformable Convolutional Network

Statement

• The main.py from MorvanZhou, and I change some code. His videos help me a lot, thank him!
• The conv_offset2D.py from oeway, and I change some code. Through his code, it is easy for me to understand deformable convolution, thank him!
• The deform_conv_my.py is transformed from the tensorflow version code of maestrojeong to pytorch, but it's still something wrong in my code.

main.py

In this code, I use a simple convolutional network and a simple deformable convolutional network to do the mnist jobs.

deformconv_pytorch's Issues

Offset output channel

DeformConv_pytorch/lib/conv_offset2D.py

Line 36 in fb3ea98

super(ConvOffset2D, self).__init__(self.filters, self.filters * 2, 3, padding=1, bias=False, **kwargs)

Your implementation is not the same as the original paper described in section 2.1. Offset output channel dimension should be 2 x N, where N is the conv kernel size k*k in 2D case. That is, the offset is shared across the channel dimension. You could refer to the implementation of https://github.com/ChunhuanLin/deform_conv_pytorch