PyTorch Implementation of Deformable Convolution

This repository implements the defromable convolution architecture proposed in this paper:
Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu and Yichen Wei. Deformable Convolutional Networks. arXiv preprint arXiv:1703.06211, 2017.

Usage

• The defromable convolution module, i.e., DeformConv2D, is defined in deform_conv.py.
• A simple demo is shown in demo.py, it's easy to interpolate the DeformConv2D module into your own networks.

TODO

• Memory effeicent implementation.
• Test against MXNet's official implementation.
• Visualize offsets
• Demo for RFCN implemantation

Notes

• Although there has already been some implementations, such as PyTorch/TensorFlow, they seem to have some problems as discussed here.
• In my opinion, the DeformConv2D module is better added to top of higher-level features for the sake of better learning the offsets. More experiments are needed to validate this conjecture.
• This repo has been verified by comparing with the official MXNet implementation, as showed in test_against_mxnet.ipynb.

Requirements

• PyTorch-v0.3.0