Scribbles4All
Code and Datasets for the NeurIPS24 Paper "Scribbles for All: Benchmarking Scribble Supervised Segmentation Across Datasets"
Install / Use
/learn @wbkit/Scribbles4AllREADME
Scribbles4All
This is the GitHub repository for the NeurIPS spotilight Paper Scribbles for All: Benchmarking Scribble Supervised Segmentation Across Datasets containing the s4Datasets and the universal scribble generator code.
Roadmap
- [x] Publish the s4 datasets
- [x] Move dataset location to GitHub
- [x] Publish the scribbble generation code
- [x] Add Croissant metadata standard
- [x] Provide download helper tools
Dataset extraction
The s4Pascal dataset can be extracted directly by calling
tar -xzf s4Pascal.tar.gz.
For the other datasets we recommend calling the respective unpacking scripts. Those will unpack the multiple tarballs and create the proper folder structure to copy-paste the scribble labels into the maind-dataset folder structure.
bash ./extractADE.sh
bash ./extractCityscapes.sh
bash ./extractKITTI360.sh
Scribble Generation/Modification
The code to generate scribble labels from fully supervised segmentation labels or modify existing scribble datasets can be found in the subfolder s4ScribbleGenerator. To get started, refer to the separate README.
License
The code in this repository an the datasets linked by this repository are published under the CC BY 4.0 license.
