Hi!
I'd like to create submissions for methods that make use of more than 2 frames: Example
In particular, imagine a method using several "support' frames before and after the reference frame for which scene flow is to be estimated.
The problem is that validating the number of frames fails for such methods:
python av2_scene_flow_competition_submit.py /dataset/av2/test_liu_flow_feather/sequence_len_003
Leads to "missing" frames in the submission.
AssertionError: COUNT MISMATCH: Found 154 files, expected 155 for sequence 0c6e62d7-bdfa-3061-8d3d-03b13aa21f68. Looking with pattern /dataset/av2/test_liu_flow_feather/sequence_len_003/0c6e62d7-bdfa-3061-8d3d-03b13aa21f68**.feather
Is there anything I would need to consider to make evaluation happen on the test set?
Locally, with some hacking and skipping the relevant frames at the beginning / end of sequences running eval on the validation set is possible of course.
Thank you!
Hi!
I'd like to create submissions for methods that make use of more than 2 frames: Example
In particular, imagine a method using several "support' frames before and after the reference frame for which scene flow is to be estimated.
The problem is that validating the number of frames fails for such methods:
python av2_scene_flow_competition_submit.py /dataset/av2/test_liu_flow_feather/sequence_len_003Leads to "missing" frames in the submission.
Is there anything I would need to consider to make evaluation happen on the test set?
Locally, with some hacking and skipping the relevant frames at the beginning / end of sequences running eval on the validation set is possible of course.
Thank you!