Function recall_precision in metrics.map wrong

I think lines

        if len(gt_boxes) > 0:
            overlaps = get_overlaps(gt_boxes, box)

            max_overlap = np.max(overlaps)
            jmax = np.argmax(overlaps)

        if max_overlap >= iou_threshold:
            if gt_checked[jmax] == 0:
                tp[prediction_index] = 1.0
                gt_checked[jmax] = 1
            else:
                fp[prediction_index] = 1.0
        else:
            fp[prediction_index] = 1.0

in metrics.map.recall_precision are wrong (link). Compare e.g. to this function from another repo.

If the GT bbox with highest overlap with the current predicted box is already matched with another predicted bbox, the current prediction bbox is marked as a false positive. However, I think, if the GT bbox with 2nd highest overlap has an overlap above the IoU threshold, then the current predicted bbox should still be marked as a true positive. Hence, instead of considering only the GT bbox with highest overlap, I think all GT bboxes with overlap above the IoU threshold should be considered. Unless I'm missing something?

So I think it should be something like this.

        if len(gt_boxes) > 0:
            overlaps = get_overlaps(gt_boxes, box)

            for overlap_idx, overlap in enumerate(overlaps):
                if gt_checked[overlap_idx] == 0 and overlap > max_overlap:
                    max_overlap = overlap
                    jmax = overlap_idx

        if max_overlap >= iou_threshold:
            if gt_checked[jmax] == 0:
                tp[prediction_index] = 1.0
                gt_checked[jmax] = 1
            else:
                fp[prediction_index] = 1.0
        else:
            fp[prediction_index] = 1.0

I'll make a PR.

ternaus / iglovikov_helper_functions

Function recall_precision in metrics.map wrong #103