antonmeskildsen / annotatr

Annotatr is a tool for creating geometric annotations for videos

Annotatr

Annotatr is a tool for creating geometric feature annotations for videos. It was created as part of a project about detecting eye features using deep neural networks and as such is still very much a work-in-progress. Right now the most important features are

• Loading/saving annotations to JSON-files
• Editing annotations using either the keyboard or direct input
• Each annotation is organised in "tracks" describing a specific property/feature for the duration of the video
• Currently points, ellipses, and rectangles are supported
• Basic copying functionality
• Easy fine-grained control of the video on a single-frame basis

Quick start

Installation

1. Clone this repository to your computer
2. Open a terminal and go to the 'client' directory
3. Run npm install
   1. Just accept default settings for Semantic UI
4. Cd into the semantic directory
5. Run gulp build

Starting the application

1. Open a terminal in the 'client' directory and type yarn start
2. Open a terminal in the 'server' directory and type the following
   1. export FLASK_APP=server.py
   2. flask run
3. The application is now ready at localhost:3000

Trackfiles

The editor uses so-called trackfiles for storing the annotations. It's just JSON and has the following form:

{
    "length": 5000,
    "video_resolution": {
        "width": 720,
        "height": 1280
    },
    "tracks": [
        {
            "name": "eye_left",
            "type": "inscribed_circle",
            "data": [
                
            ]
        }
    ]
}

Doing more with your data

Processing is a data processing framework developed specifically for computer vision purposes. It supports a wide array of transformations on trackfiles as well as videos and images and is purpose-built for easy dataset creation.