Code for the paper:
Efficient Image Annotation via Semi-Supervised Object Segmentation with Label Propagation Vitalii Tutevych, Raphael Memmesheimer, Luca Eichler, Dmytro Pavlichenko, Fynn Schilke, Rodja Krudewig, Sven Behnke https://arxiv.org/abs/2604.22992
The dataset used in the paper can be downloaded here:
Place the extracted contents under datasets/ so each object class has its own subdirectory with train/ and valid/ splits.
pip install -r requirements.txtPlace your YAML dataset recipes in recipes/ and raw datasets in datasets/.
Set THEIA_CACHE_DIR to your local Theia model directory if not using the default (~/.cache/theia/).
Train labelers (trains CLIP, Theia, and ViT Hopfield classifiers on representative samples):
python scripts/train_labelers.pyPropagate labels to a validation set and evaluate:
python scripts/validate_labelers.pyLabel raw datasets using a detection model (requires ultralytics):
python scripts/segment_proposer.py --to_label_dir datasets/ --detector_dir /path/to/model