mmdetection遇到的奇奇怪怪的问题(遇到bug就更新)

2.2 'ConvWS is already registered in conv layer'

https://github.com/open-mmlab/mmdetection/issues/3249

2.1 eval时bbox全部为空

玄学问题，一般只要在data中把classes加上就可以了

asset CUDA device

大概率是num_classes设置不对, coco80类, 1x是设置81，2x是设置80

RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. This error indicates that your module has parameters that were not used in producing loss

我自定义的后端加上fp16训练正常，去掉就会报这个error
按照 https://github.com/open-mmlab/mmdetection/issues/2153 改apis/train.py可以解决
我的解决是config加上

find_unused_parameters = True

1.2 中的iou loss报错

follow https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/losses/iou_loss.py#L139
将weights的维度降到1就可以了

2.2中自定义的coco数据集加载为空

取消设置classes 手动修改coco.py中的class和class_names.py中的class

2.0中获取不到gt_labels

需要手动指定classes=classes