Weighting Pseudo-Labels via High-Activation Feature Index Similarity and Object Detection for Semi-Supervised Segmentation [ECCV 2024]

Contact

If you have any questions, please email Prantik Howlader at phowlader@cs.stonybrook.edu.

Dataset

• Pascal: JPEGImages | SegmentationClass
• Cityscapes: leftImg8bit | gtFine

Please modify your dataset path in configuration files.

├── [Your Pascal Path]
    ├── JPEGImages
    └── SegmentationClass
    
├── [Your Cityscapes Path]
    ├── leftImg8bit
    └── gtFine
    

For generating bounding boxes for training segmentation network, check fasterrcnn for instructions

Usage

UniMatch + Ours

# use torch.distributed.launch
sh scripts/train.sh

Citation

If you find this project useful, please consider citing:

@article{howlader2024weighting,
  title={Weighting Pseudo-Labels via High-Activation Feature Index Similarity and Object Detection for Semi-Supervised Segmentation},
  author={Howlader, Prantik and Le, Hieu and Samaras, Dimitris},
  journal={arXiv preprint arXiv:2407.12630},
  year={2024}
}