This repository provides source code for the MICCAI2023 paper "Adaptive Region Selection for Active Learning in Whole Slide Image Semantic Segmentation" [arXiv
] [blogpost
].
You may set "sampling_strategy" in user_define.py:
- "full" (full annotation benchmark)
- "random"
- "uncertainty_standard"
- "uncertainty_non_square"
- "uncertainty_adapt"
You may set "n_query" and "region_size" in user_define.py to define AL parameters, and "CYCLES" in experiments.py to define the number of conducted AL cycles.
python experiments.py
We use Fastai_v1 for implementation. We observed some package conflicts while building the conda env, you may install as the following:
conda create -n fastai_v1_38 python=3.8
source activate fastai_v1_38
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
conda install -c fastai fastai=1.0.61
conda install pip
pip install -r requirements.txt
conda install pixman=0.40.0
conda install -c conda-forge openslide
We trained on the fully-annotated data to validate our segmentation framework. In full annotation benchmark, you may find two trained models.