the length of scenario.timestamps_seconds is not equal to tracks[i].states_size()

[changjiang8]

When processing Motion Dataset (training_20s&prefix=&forceOnObjectsSortingFiltering=false)), we found that the length of timestamp_seconds of some scenarios is not equal to tracks[i].states_size() for all tracks. A simple analysis found that the timestamps of some scenarios will be lost, and the Track.states of other scenarios will be lost; How to deal with this situation? We now directly eliminate this part of the Scenario.

A total of 135 scenarios were found in 124 files with the exception mentioned above.

Case of missing timestamp:

scenario_id 33e5ac4c35a5af28 training_20s.tfrecord-00123-of-01000 len(timestamps_seconds) = 193 len(track.states) = 199

Case of missing states:

scenario_id c34dc63d55ca05f5 training_20s.tfrecord-00313-of-01000 len(timestamps_seconds) = 199 len(track.states) = 196

[scott-ettinger]

Thanks for raising this. We will take a look into the cause. I think for now there are two possible options. Since this affects only a tiny fraction of the data, your solution of truncating the affected scenarios is a reasonable method. I would think it should not significantly affect overall results of experiments. The other option would be to extend the timestamps by assuming a nominal 0.1 second interval per timestamp. This would approximate the timestamps with some small error.

[changjiang8]

Thanks for your reply. I think the second option is good, so I will try it to reuse the scenarios with missing timestamps.