Considering match before last frameid when re-establish tracks will lead to unpredicted match and neglected id tranfer.
From result on private dataset and result, HId 148 trasfer bettween 23 and 27, but not counted timely.
Before bug fix:
events.loc[event["HId"]==148]
events.loc[event["0Id"]==23]
events.loc[event["0Id"]==27]
After bug fix:
events.loc[event["HId"]==148]
iou_det_SDP.zip
When testing the iou_det_SDP.zip on MOT16_train, average MOTA is 62.9%(pymotmetrics) with 63.1%(mot challendge matlab toolbox)
By the way, ML MT definition is slight different with matlab toolbox, but it's unimportant:
pymotmetrics: ML >= 0.8; partially_tracked 0.2 <= <0.8
matlab: ML > 0.8; partially_tracked 0.2 <= <=0.8
If change the ML MT definition in metrics.py as well, result will be:
The commit is tested in master branch, but mot.py is same in the branch.
cheind
commented
11 months ago
Considering match before last frameid when re-establish tracks will lead to unpredicted match and neglected id tranfer.
From result on private dataset and result, HId 148 trasfer bettween 23 and 27, but not counted timely. Before bug fix: events.loc[event["HId"]==148]
events.loc[event["0Id"]==23]
events.loc[event["0Id"]==27]
After bug fix: events.loc[event["HId"]==148]
iou_det_SDP.zip When testing the iou_det_SDP.zip on MOT16_train, average MOTA is 62.9%(pymotmetrics) with 63.1%(mot challendge matlab toolbox)
By the way, ML MT definition is slight different with matlab toolbox, but it's unimportant: pymotmetrics: ML >= 0.8; partially_tracked 0.2 <= <0.8 matlab: ML > 0.8; partially_tracked 0.2 <= <=0.8
If change the ML MT definition in metrics.py as well, result will be:
The commit is tested in master branch, but mot.py is same in the branch.