Performance not competitive for single class detection.

WongKinYiu / yolov7

Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

GNU General Public License v3.0

13.36k stars 4.22k forks source link

Performance not competitive for single class detection. #87

Open jocarias opened 2 years ago

jocarias commented 2 years ago

When there is only a single class to detect, we are experiencing a lack of performance when compared with YOLOv4 (darknet) and YOLOv5. We experienced a similar situation with YOLOR.

Besides changing the nc settings in "coco.yaml" file, is there additional changes we should do for the single class cases?

mnurilmi commented 2 years ago

@jocarias is this problem clear? I mean, I'm having the same problem...

jocarias commented 2 years ago

At the moment I've been using YOLOv5 so I don't know the state of the problem.

To be clear, I was training the pre-trained weights for a new class.

mnurilmi commented 2 years ago

Ok, thanks for the answer