voc数据集训练totallossnan lossnan问题？

[ChenMaolong]

大佬，我用的model_path = 'model_data/yolo_weights.pth'跑的voc单类别数据集，类别也改了， Start Validation Epoch 73/600: 100%|██████████| 14/14 [00:08<00:00, 1.66it/s, val_loss=nan] Finish Validation

Intel MKL ERROR: Parameter 6 was incorrect on entry to DGELSD. Epoch:73/600 Total Loss: nan || Val Loss: nan Start Train Epoch 74/600: 100%|██████████| 118/118 [00:28<00:00, 4.10it/s, loss=nan, lr=0.000965] Finish Train Start Validation Epoch 74/600: 100%|██████████| 14/14 [00:08<00:00, 1.64it/s, val_loss=nan] Finish Validation

Intel MKL ERROR: Parameter 6 was incorrect on entry to DGELSD. 出现这种问题？是因为预训练权重的原因吗但是你github只有coco的权重有voc的吗？

[bubbliiiing]

感觉是环境配置的问题

[ChenMaolong]

嗯嗯，我试试