search

zzh8829 / yolov3-tf2

YoloV3 Implemented in Tensorflow 2.0
MIT License
2.51k stars 913 forks source link

Correct score for one-class detection #380

Closed makra89 closed 2 years ago

makra89 commented 2 years ago

See comment in See https://github.com/zzh8829/yolov3-tf2/pull/311

I think the score in yolo_nms is calculated wrong when only having one class. The model cannot learn anything about the class since the associated loss is always 0 --> class probability is always <= 0.5 --> score is always <= 0.5

Ah and by the way: Awesome work :) Your model works very well when trained trying to build an aimbot: https://github.com/makra89/Quake_AI

MatthewPerryman commented 2 years ago

I don't know if it's commonplace to write on these, but with this change my one class object detector works perfectly thanks!