Multi-Stream Convolutional Neural Network for Action Recognition Based on Adaptive Visual Rhythm

This repository contains some files that were taken originally from Two Stream Pytorch.

The main contribution focuses on the creation of a new stream (convolutional neural network) that is trained with different type of visual rhythm images generated from videos of UCF101 and HMDB51 datasets.

Installation

Tesded on PyTorch

OS: Ubuntu 16.04
Python: 3.5
CUDA: 9.0.176
OpenCV3
dense_flow

Some important tips to successfully install dense_flow are shown in the aforementioned link.

Code also work for Python 2.7.