def matchGtWithPred()

in maskrcnn_benchmark/data/datasets/evaluation/cityscapes/eval_instances.py [0:0]
30 lines of code
11 McCabe index (conditional complexity)

def matchGtWithPred(dataset, predictions, idx):
    # Collect instances from gt and pred separately per image
    # TODO: not parallel! parallelize this process safely
    perImgGtInstances, gtMasks = prepareGtImage(dataset, idx)
    perImgPredInstances, predMasks = preparePredImage(dataset, predictions, idx)

    # If no masks are provided, the segmentation score will be 0
    for gt, gtMask in zip(perImgGtInstances, gtMasks):
        for pred, predMask in zip(perImgPredInstances, predMasks):
            if not isOverlapping(gt["box"], pred["box"]):
                continue

            boxIntersection = computeBoxIntersection(gt, pred)
            maskIntersection = computeMaskIntersection(gt, gtMask, pred, predMask)

            if boxIntersection > 0:
                # Copy metadata only, and register the matched pairs
                # this step is redundant but informative
                # intersection score would be enough
                gtCopy = gt.copy()
                predCopy = pred.copy()

                # remove linking field (an empty list) to avoid confusion
                gtCopy.pop("matchedPred")
                predCopy.pop("matchedGt")

                gtCopy["boxIntersection"] = boxIntersection
                gtCopy["maskIntersection"] = maskIntersection
                predCopy["boxIntersection"] = boxIntersection
                predCopy["maskIntersection"] = maskIntersection

                gt["matchedPred"].append(predCopy)
                pred["matchedGt"].append(gtCopy)

    # Group by classes
    groupedGtInstances = {labelName: [] for labelName in dataset.CLASSES}
    groupedPredInstances = {labelName: [] for labelName in dataset.CLASSES}

    for gt in perImgGtInstances:
        gtLabelName = dataset.id_to_name[gt["labelID"]]
        groupedGtInstances[gtLabelName].append(gt)

    for pred in perImgPredInstances:
        predLabelName = dataset.id_to_name[pred["labelID"]]
        groupedPredInstances[predLabelName].append(pred)

    match = {"groundTruth": groupedGtInstances, "prediction": groupedPredInstances}

    return match