Soft Teacher error

[Aniulo]

1. 修改不同 batch_size尝试后，还是出现如下错误：

   losses['loss_cls'] = losses['loss_cls'] * len( KeyError: 'loss_cls'

2. 在一次log里训练没有报错，是我手动停止的，原因是显示剩余训练时间为负数 2024/05/22 14:38:27 - mmengine - INFO - Epoch(train) [1][452650/50] lr: 1.0000e-02 **_eta: -2 days, 20:22:30_** time: 0.2106 data_time: 0.0033 memory: 4333 loss: 0.0344 sup_loss_rpn_cls: 0.0039 sup_loss_rpn_bbox: 0.0021 sup_loss_cls: 0.0135 sup_acc: 99.4141 sup_loss_bbox: 0.0133 unsup_loss_rpn_cls: 0.0003 unsup_loss_rpn_bbox: 0.0000 unsup_loss_cls: 0.0012 unsup_acc: 100.0000 unsup_loss_bbox: 0.0000

[nimeishibaaa]

same first error

[XiaoSiJi-GCY]

It seems that Soft Teacher can only use "IterBasedTrainLoop" and set "by_epoch=False"