PyTorch Custom Convolution Layer

This project is about how to define a custom convolution layer in PyTorch, and use CUDA function to implement convolution.

Content

/cpp C++ extension include CUDA interface and Python module bind.

/cuda Implicit gemm convolution implementation.

/include Declaration about forward/backward convolution.

/pytorch Include setup.py script, custom convolution layer definition.

Build

$ sh setup.sh

If you don't have root permission, add environment option --prefix="/home/user/.conda/envs/yourenvname/.

TEST

$ cd pytorch
vim test.py

RUN

run inference program

$ python inference.py