README.md

MULTI-DIRECTIONAL CONVOLUTION

1 Paper Details

B. Yang, Z. Wang, W. Ran, H. Lu and Y. -P. P. Chen, "Multi-Directional Convolution Networks with Spatial-Temporal Feature Pyramid Module for Action Recognition," ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 2440-2444, doi: 10.1109/ICASSP39728.2021.9413610. [ICASSP 2021]

The following pictures are examples of spatial-temporal motion decomposition and residual block with MDConv,respectively.

2 Usage

2.1 Install C++ Extension

cd mdconv
python setup.py install

2.2 Import Package

import mdconv

2.3 Example

# You can copy MDConvFunction, DirectionalConv and MDConv, and use MDConv instead of Conv3d.
self.conv = MDConv(in_channels, out_channels, stride=1, bias=False)
# forward
out = self.conv(x)