Introduction

This repo is a pytorch implementation of Deformable Convolution Networks. It is ported from a previous pytorch implemetnation, which is transformed from original MXNet implementation.

What's the difference from other implementations?

This repo supports Pytorch 1.0, which uses a more convinient C++ and Cuda exntesion tools ATen.

Todo list

• [ ] Found a minor bug in pytorch 1.0 version (do not use 1.0 for now).
• [ ] Benchmark the performance on object detection (faster r-cnn).

Compilation

Prerequiestes

• Python 2.7 or 3.6
• Pytorch 0.4 or Pytorch 1.0
• CUDA 8.0 or higher

Build Pytorch-0.4 version

sh make.sh
CC=g++ python build.py

Build Pytorch-1.0 version

python setup.py build develop

See test.py for example usage.

Notice

Only torch.cuda.FloatTensor is supported.