dab-detr train AP=0

[bye111]

问题确认 Search before asking

• [X] 我已经搜索过问题，但是没有找到解答。I have searched the question and found no related answer.

请提出你的问题 Please ask your question

而且训练的损失好多大于1：base_lr": 0.0001, "lr": 0.0001, "data_time": 0.006839370727539063, "grad_norm": 49.24021549224854, "loss": 25.259742126464843, "loss_cls": 0.8597541677951813, "loss_bbox": 1.51627574801445, "loss_iou": 1.838999490737915, "d0.loss_cls": 0.8601799380779266, "d0.loss_bbox": 1.49742307305336, "d0.loss_iou": 1.8469653284549714, "d1.loss_cls": 0.8606949377059937, "d1.loss_bbox": 1.5037752544879914, "d1.loss_iou": 1.8431547117233276, "d2.loss_cls": 0.8608416128158569, "d2.loss_bbox": 1.5059378218650818, "d2.loss_iou": 1.8428979802131653, "d3.loss_cls": 0.8595381605625153, "d3.loss_bbox": 1.505025475025177, "d3.loss_iou": 1.8457740879058837, "d4.loss_cls": 0.8594501793384552, "d4.loss_bbox": 1.5172400069236756, "d4.loss_iou": 1.835814266204834, "time": 0.44253116130828857, "epoch": 10, "iter": 20614, "memory": 3001, "step": 20614} 目前尝试num_classes+1也没改变，我使用的预训练模型为deformable_detr_twostage_refine_r50_16x2_50e_coco_20210419_220613-9d28ab72.pth，学习率为0.0001，请问是哪里的问题，训练cascade rcnn系列都没有问题

[lyuwenyu]

deformable_detr_twostage_refine_r50_16x2_50e_coco_20210419_220613-9d28ab72.pth 这是torch的？

[188080501]

同样的问题 我训练DETR-R50模型，全部都是0，跑推理啥都识别不到 同样的训练数据，换PP-YOLOE_plus-L模型就没问题