search

SysCV / qdtrack

Quasi-Dense Similarity Learning for Multiple Object Tracking, CVPR 2021 (Oral)
Apache License 2.0
384 stars 61 forks source link

Training accuracy meaning #96

Closed AmanGoyal99 closed 2 years ago

AmanGoyal99 commented 3 years ago

I had put the model on training for BDD100K dataset. I wanted to know what is the accuracy metric which appears after every epoch when model is trained on BDD100K? Is it MOTA, BDD100K or anything else ? Kindly have a look at the below snippet to have a better understanding of which accuracy I am talking about.

2021-11-03 10:48:23,006 - qdtrack - INFO - Epoch [6][89200/171715] lr: 2.000e-02, eta: 5 days, 16:56:06, time: 0.441, data_time: 0.012, memory: 7072, loss_rpn_cls: 0.0493, loss_rpn_bbox: 0.0109, loss_cls: 0.2063, acc: 92.8965, loss_bbox: 0.2769, loss_track: 0.0937, loss_track_aux: 0.0646, loss: 0.7016

I wanted to know what does the accuracy term indicate here ? Is It MOTA, IDF1, etc ?

OceanPang commented 2 years ago

It's a classification acc from the detector.