Synthetic video segmentation dataset used for object-centric representation learning.
This repository contains the code used to create the dataset. A link to the final dataset will be included once created.
Download the following folders from the WebGLSamples repository https://github.com/WebGLSamples/WebGLSamples.github.io.
aquarium
jquery-ui-1.8.2.custom
khronos
tdl
Put them in the basedir of this repository. Now apply the patch to the files in the aquarium folder.
git apply aquarium/patch
Now install the dependencies.
npm install
To run the aquarium, use a static webserver to serve from the repositories root directory and point your web browser to the respective url:
http-server .
To record a dataset, you have to
-
Generate a set of random sample configurations:
node generate_random_aquarium_configs.js <path> <count>
-
Start workers to record the samples:
node server.js <path>
You can start multiple workers for the same sample set. Once finished, the recorded datasets is located at
<path>/final
.
This work is adapted from a WebGL demo obtained from GitHub. The original demo is licensed under the 3-clause BSD license, a copy of which is included as THIRD_PARTY_LICENSE.md.