sgherbst / cs229-project

Real-time analysis of fruit-fly videos

Introduction

This project is a tool for real-time video analysis of fruit flies, and is capable of determining fly position, orientation, sex, and wing angle (for male flies).

Installation

After cloning this repository, navigate to the top-level directory, create a new conda environment and use pip to install the project and its dependencies:

> conda create --name=cs229-project python=3.6
> source activate cs229-project
> pip install -e .

If you choose a name other than cs229-project, please update the PROJECT_NAME variable in the toplevel Makefile accordingly. Similarly, if you are not using miniconda, please update the definition of the Makefile variable CONDAROOT.

Then download the the input directory (2 GB) and place it in the top-level directory. This folder, containing two subfolders images and video, contains the raw and labeled data used for this project.

Running the examples

In the top-level directory, you can train the all four models used as part of the image processing pipeline by simply running

> make

This should take about a minute. After the models are trained, you can view the video processing in realtime by running

> make demo

To run the video processing on other video clips, navigate to the cs229 directory and run

> python demo.py -i testN.mp4

where N is replaced with a value from 1 to 5 (the default is 4).

Finally, to run a performance test (which processes the video but does not display the results live), return to the top-level directory and run:

> make profile