Error when running camera_segmentation_utils.decode_multi_frame_panoptic_labels_from_segmentation_labels: max() arg is an empty sequence

[PradeepKadubandi]

For some training data, the above method causes the error as mentioned.

The sample context and sequence for which this happens: context: 10724020115992582208_7660_400_7680_400 sequence: 15672339233166359051

The issue happens because all the protos in the above sequence have empty lists for local and global ids; in such a case, the code attempts to call max with an empty sequence.

This is the call stack (the relevant parts) if helpful:

    panoptic_labels, is_tracked_masks, num_cameras_covered, panoptic_label_divisor = camera_segmentation_utils.decode_multi_frame_panoptic_labels_from_segmentation_labels(
  File "/home/pkadubandi/.local/share/virtualenvs/waymo-challenge-JtwqplnD/lib/python3.9/site-packages/waymo_open_dataset/utils/camera_segmentation_utils.py", line 292, in decode_multi_frame_panoptic_labels_from_segmentation_labels
    global_id_mapping = _remap_global_ids(
  File "/home/pkadubandi/.local/share/virtualenvs/waymo-challenge-JtwqplnD/lib/python3.9/site-packages/waymo_open_dataset/utils/camera_segmentation_utils.py", line 177, in _remap_global_ids
    max_instance_id = max(global_instance_ids[sequence])
ValueError: max() arg is an empty sequence

[alexzzhu]

Thanks for pointing this out. The below code snippet should hopefully fix this, please let me know if you run into further issues. I'll push this to GitHub asap.

cumulative_max_instance_id = 0
global_instance_ids_offset = {}
for sequence in global_instance_ids:
  if not global_instance_ids[sequence]:
    continue
  max_instance_id = max(global_instance_ids[sequence])
  global_instance_ids_offset[sequence] = [
      inst_id + cumulative_max_instance_id
      for inst_id in global_instance_ids[sequence]
  ]
  cumulative_max_instance_id += max_instance_id

if not global_instance_ids_offset:
  return {}