I3D models trained on Kinetics

Overview

For this project, I am building an end-to-end trainable behavior recognition system for mice using deep convolutional networks. These networks are inspired from Inception-3d, the current state-of-the-art in video action recognition. Please find detailed information about this architecture in the paper "Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset" by Joao Carreira and Andrew Zisserman. The paper was posted on arXiv in May 2017, and was published as a CVPR 2017 conference paper. Below is an architecture diagram of Inception-3D.

Running the code

Setup

1. Follow the instructions for installing Sonnet.

2. clone this repository using

$ git clone https://github.com/vijayvee/behavior_recognition

3. Add the cloned repository's parent path to $PYTHONPATH as follows

cd <parent_dir>/behavior_recognition; export PYTHONPATH=$PYTHONPATH:<parent_dir>

Acknowledgments

• The Kinetics dataset
• Inception v1
• Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset