Formatting Object Detection Labels - Multiple Predictions but Empty Labels

[jeethesh-pai]

Hi CleanLab team,

Thanks a lot for this repo. I would like to know how should we format the Object detection label for label error detection in cleanlab.

According to tutorial, it is just mentioned that the labels[i] should be a dictionary with "labels" and "bboxes" as keys. How should we handle the case for no label images. Image with just background. I am getting an error because of this issue.

    label_issue_idx = find_label_issues(labels, predictions, return_indices_ranked_by_score=True)
  File "/usr/local/lib/python3.8/dist-packages/cleanlab/object_detection/filter.py", line 115, in find_label_issues
    is_issue = _find_label_issues(
  File "/usr/local/lib/python3.8/dist-packages/cleanlab/object_detection/filter.py", line 138, in _find_label_issues
    per_class_scores = _get_per_class_ap(labels, predictions)
  File "/usr/local/lib/python3.8/dist-packages/cleanlab/object_detection/filter.py", line 398, in _get_per_class_ap
    ap_per_class = _calculate_ap_per_class(labels, predictions, iou_threshold=threshold)
  File "/usr/local/lib/python3.8/dist-packages/cleanlab/object_detection/filter.py", line 275, in _calculate_ap_per_class
    pred_bboxes, lab_bboxes = _filter_by_class(labels, predictions, class_num)
  File "/usr/local/lib/python3.8/dist-packages/cleanlab/object_detection/filter.py", line 317, in _filter_by_class
    lab_bboxes.append(label["bboxes"][gt_inds, :])
IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed

This error arises because some of images have empty labels (just background) and labels looks like this,
label = {"label": array([]), "bboxes": array([])}

Any help regarding the same would be greatly appreciated.

Thanks and regards
Jeethesh

[jwmueller]

This should be addressed once this PR is in: #1235