Repository navigation
why My YOLOX tend to output False positive result, it's really hurting my tracking performance, Can any body help? #448
Description
when i was try to reproduce the tracking result of bytetrack, I found my detector tend to output false positive bounding box. so when i was upload my tracking result in motchanllenge, my FN is around 69000, but my FP is around 50000, it's so strange, cause i had set the init_new_track score threshold to 0.8, it still have 50000 FP, i had review the code and all setting is the same with ByteTrack, even some code was directly copy from here, so curious about this.
Activity
iamxzl0314 commentedon May 25, 2025
of cource. my Yolox weight is directly download from here model zoo, and i also used the corresponding dataset weight, but still had so much FP.
iamxzl0314 commentedon May 25, 2025
such as if we use 0.6 as the init_new_track threshold, in my Yolox detction result there will be 27 bounding box in the first frame of the first test vedio, but i had download many tow-stage trackers tracking result from the motchanllenge webside, they are almost share the same tracks management logist, and also use YOLOX as their detector, but all of them including ByteTrack, just get 13-15 bounding boxes. why so different?