mAP = 0 all the time
Kangzf1996 opened this issue · 7 comments
In the training log, the mAP is 0 all the time
bbox_mAP: 0.0000, bbox_mAP_50: 0.0000, bbox_mAP_75: 0.0000, bbox_mAP_s: -1.0000, bbox_mAP_m: -1.0000, bbox_mAP_l: 0.0000, bbox_mAP_copypaste: 0.000 0.000 0.000 -1.000 -1.000 0.000, segm_mAP: 0.0000, segm_mAP_50: 0.0000, segm_mAP_75: 0.0000, segm_mAP_s: -1.0000, segm_mAP_m: -1.0000, segm_mAP_l: 0.0000, segm_mAP_copypaste: 0.000 0.000 0.000 -1.000 -1.000 0.000
Do you know how to solve it? Thank you
Do you train QueryInst on COCO?
Do you train QueryInst on COCO?
thank you for your reply.
No, I train it on my dataset. And I think the dataset is ok because I trained maskrcnn on it sucessfully.
I suggest you can try to tune the lr, warmup lr and other hyperparameters. Can you successfully train Sparse R-CNN on your dataset?
I suggest you can try to tune the lr, warmup lr and other hyperparameters. Can you successfully train Sparse R-CNN on your dataset?
I didn't try with Sparse R-CNN. For now, the lr is 0.000025, maybe it's too little?
I suggest you try to tune Sparse R-CNN first, the query-based approaches are much less robust than Faster / Mask R-CNN in term of training and optimization.
I suggest you try to tune Sparse R-CNN first, the query-based approaches are much less robust than Faster / Mask R-CNN in term of training and optimization.
ok, I will try to tune the lr and try Sparse R-CNN. Thank you for your suggestion.
你好，请问你解决这个问题了吗