Request for additional performance metrics
jxmmy7777 opened this issue · 1 comments
Hello @stepankonev ,
Thank you for sharing your code on Github. I have been reviewing the technical report and have noticed that only the mAP on the test results is reported. I was wondering if it would be possible for you to share other performance metrics such as minADE, minFDE, etc. Additionally, I was wondering if the performance of the model seems to be consistent with the validation set/test set.
Thank you for your time and assistance. I look forward to hearing from you.
Best regards,
Hello @jxmmy7777 ,
Thank you for your attention to the repository. Test results can be found at the official competition website under MPA name and currently are the following:
Object Type | Measurement Time (s) | Soft mAP | mAP | Min ADE | Min FDE | Miss Rate | Overlap Rate |
---|---|---|---|---|---|---|---|
Vehicle | 3 | 0.5656 | 0.5497 | 0.2914 | 0.5332 | 0.0931 | 0.0226 |
Vehicle | 5 | 0.4469 | 0.4408 | 0.6099 | 1.2239 | 0.1429 | 0.0491 |
Vehicle | 8 | 0.3276 | 0.3242 | 1.2022 | 2.7640 | 0.2256 | 0.1064 |
Vehicle | Avg | 0.4467 | 0.4382 | 0.7012 | 1.5070 | 0.1539 | 0.0594 |
Pedestrian | 3 | 0.4704 | 0.4606 | 0.1697 | 0.3122 | 0.0716 | 0.2405 |
Pedestrian | 5 | 0.3708 | 0.3646 | 0.3246 | 0.6584 | 0.1040 | 0.2641 |
Pedestrian | 8 | 0.3081 | 0.3023 | 0.5771 | 1.2843 | 0.1359 | 0.2948 |
Pedestrian | Avg | 0.3831 | 0.3758 | 0.3572 | 0.7516 | 0.1038 | 0.2665 |
Cyclist | 3 | 0.4265 | 0.4216 | 0.3370 | 0.6292 | 0.1919 | 0.0460 |
Cyclist | 5 | 0.3629 | 0.3596 | 0.6459 | 1.2922 | 0.2135 | 0.0880 |
Cyclist | 8 | 0.2585 | 0.2563 | 1.1636 | 2.5588 | 0.2640 | 0.1401 |
Cyclist | Avg | 0.3493 | 0.3458 | 0.7155 | 1.4934 | 0.2231 | 0.0914 |
Avg | 3 | 0.4875 | 0.4773 | 0.2660 | 0.4915 | 0.1189 | 0.1030 |
Avg | 5 | 0.3935 | 0.3883 | 0.5268 | 1.0582 | 0.1534 | 0.1338 |
Avg | 8 | 0.2981 | 0.2943 | 0.9810 | 2.2024 | 0.2085 | 0.1804 |
Avg | Avg | 0.3930 | 0.3866 | 0.5913 | 1.2507 | 0.1603 | 0.1391 |
No significant divergence between test and val metrics was observed if using same dataset version