/best_practices_ALSS

This is the official repository for the paper - Best Practices in Active Learning for Semantic Segmentation

Primary LanguagePython

best_practices_ALSS

AL for Semantic Segmentation

This part is the official repository for the paper - Best Practices in Active Learning for Semantic Segmentation

Datasets

  • PASCAL-VOC (Augmented train set: 10582 images)
  • Cityscapes
  • A2D2
    • Pool-0f
    • Pool-5f
    • Pool-11f
    • Pool-21f
    • Pool-Aug

Query Strategies for Semantic Segmentation

  • Random Sampling
  • Entropy-based Sampling
  • CoreSet Approach
  • EquAL Sampling
  • Random Sampling SSL
  • Entropy SSL
  • CoreSet SSL
  • EquAL SSL

Sample script for PASCAL-VOC

python run.py --random-image --config ./configs/datasets/pascal_voc.yaml
python run.py --entropy-image --config ./configs/datasets/pascal_voc.yaml
python run.py --coreset --config ./configs/datasets/pascal_voc.yaml

Sample script for A2D2

python run.py --random-image --config ./configs/datasets/a2d2.yaml

You can download the weights for the pretrained weights (Wide-ResNet-38) for initializing the encoder backbone here

AL for Image Classification

Datasets

  • CIFAR-10
  • CIFAR-100

Query Strategies for Image Classification

  • Random Sampling
  • Entropy Sampling
  • CoreSet
  • Learning Loss
  • Ensemble Entropy
  • Ensemble Variation Ratio (QBC)

Citation

@InProceedings{ALSS_2024_GCPR,
    author    = {Mittal, Sudhanshu and Niemeijer, Joshua and Sch{\"a}fer, J{\"o}rg P. and Brox, Thomas},
    title     = {Best Practices in Active Learning for Semantic Segmentation},
    booktitle = {Proceedings of the DAGM German Conference on Pattern Recognition},
    year      = {2023},
}