Developed for nuclei instances segmentation in 2d and 3d. Computes average precision (AP = TP/(TP+FP+FN))

The recommended way is to install it into your conda/python virtual environment.

conda activate <<your-env-name>>
git clone https://github.com/Kainmueller-Lab/evaluate-instance-segmentation
cd evaluate-instance-segmentation
pip install -e .

    python evaluate.py --res_file <pred-file> --res_key <hdf-key> --gt_file <gt-file> --gt_key <hdf-key> --out_dir output --background 0

(--res_key and --gt_key only required for hdf or zarr input)

By default hungarian matching is used to compute the matches of prediction and ground truth (can be disabled with --no_use_linear_sum_assignment.
AP is computed for multiple thresholds.

• evaluation metrics are written to toml-file and returned as dict