search

zylo117 / Yet-Another-EfficientDet-Pytorch

The pytorch re-implement of the official efficientdet with SOTA performance in real time and pretrained weights.
GNU Lesser General Public License v3.0
5.2k stars 1.27k forks source link

Validation loss is not stable for both classification and regression #707

Open alsheabi opened 2 years ago

alsheabi commented 2 years ago

Any suggestion or idea about this issue?

loss

zylo117 commented 2 years ago

category id starts from 1, 0 is the background

alsheabi commented 2 years ago

category id starts from 1, 0 is the background

@zylo117 Thank you for replying, regards category id start from 1 I did that for coco, but I don't get what do you mean about 0 for background :( do you mean segmentation in Annotation? Could you please tell me what it's mean?

cate and Annotation for Bbox looks like this: ann Or you mean IoU it's already 0

back