Boosting-R-CNN environment error

我的cuda版本是11.7，mmcv1.4版本太低，会冲突，但是安装高版本2.0.0的mmcv会报如下的错误，高版本集成方法的位置发生了变化，请问该怎么解决

Sep 21 '24 08:09 Nicoleyk

我的cuda版本是11.7，mmcv1.4版本太低，会冲突，但是安装高版本2.0.0的mmcv会报如下的错误，高版本集成方法的位置发生了变化，请问该怎么解决

cuda11.7就装不了mmcv1.4了吗。。。我也不太清楚怎么解决，我建议你慢慢从mmcv1.5 1.6 1.7慢慢往上试一试，看看哪个版本能装得上又能用的

Sep 21 '24 12:09 mousecpn

我解决好环境问题了，谢谢。但是在运行自己的数据集的过程中遇到了如下的错误 Uploading 1729046009167.png… 期待您的回复

Oct 16 '24 02:10 Nicoleyk

图好像裂开了，看不到

Oct 16 '24 07:10 mousecpn

File "/home/yk/Boosting-R-CNN-master/Boosting-R-chN-master/mmdet/models/roi heads/prob roi head.py", line 143, in box forward train boost loss bbox['loss cls'1 = self.norm loss(loss bbox['loss cls'l, label weights new, label weights new.shape[@]) File "/home/yk/Boosting-R-chN-master/Boosting-R-CNN-master/mmdet/models/roi heads/prob roi head.py", line 152, in norm lossnew weights =weights *(loss.sum()/(weights * loss).sum()) RuntimeError: The size of tensor a (2898) must match the size of tensor b (2848)at non-sineleton dimension 0

Oct 16 '24 07:10 Nicoleyk

我跑的是其他的水下数据集，在第一轮的运行过程中会报这种错

Oct 16 '24 07:10 Nicoleyk

File "/home/yk/Boosting-R-CNN-master/Boosting-R-chN-master/mmdet/models/roi heads/prob roi head.py", line 143, in box forward train boost loss bbox['loss cls'1 = self.norm loss(loss bbox['loss cls'l, label weights new, label weights new.shape[@]) File "/home/yk/Boosting-R-chN-master/Boosting-R-CNN-master/mmdet/models/roi heads/prob roi head.py", line 152, in norm lossnew weights =weights *(loss.sum()/(weights * loss).sum()) RuntimeError: The size of tensor a (2898) must match the size of tensor b (2848)at non-sineleton dimension 0

你可能是类别没改，你要去dataset那里改一改类别

Oct 16 '24 07:10 mousecpn

请问你最终用的什么版本的mmcv解决的cuda版本过高的问题

Jul 10 '25 21:07 MaKaBaKaooooo