This script is designed to evaluate the performance of a panoptic segmentation model using the Uncertainty-Aware Panoptic Quality (AUPQ) metric. This metric is an extension of the Panoptic Quality (PQ) metric that takes into account the model's uncertainty in its predictions.
python uncertainty_aware_panoptic_quality.py --gt_json_file <gt_json_file> --pred_json_file <pred_json_file> --gt_folder <gt_folder> --pred_folder <pred_folder> --nr_thresholds <nr_thresholds>
--gt_json_file: The path to the JSON file with ground truth data.--pred_json_file: The path to the JSON file with predictions data.--gt_folder: (optional) The folder with ground truth COCO format segmentations. Default: X if the corresponding json file is X.json.--pred_folder: (optional) The folder with prediction COCO format segmentations. Default: X if the corresponding json file is X.json.--nr_thresholds: (optional) The number of thresholds for uncertainty. Default is 16.
This script is based on the official COCO Panoptic Quality (PQ) evaluation script.