Open WangYuSenn opened 2 weeks ago
I tested again and did not find the issue.
Can you share me the detailed log?
Now that the weights I loaded up, is there a problem with the layering of the backbone, with base as the force, after the layer operation, the output has two bs x 1024 x 8 x 8 tensor
So can you show me the detailed log? maybe I can help you by comparing the log output with the log file in the release.
你好,我用的vmamba_base_s2l15作为backbone,我通过将vmamba.py的1520行经过layer的特征输出之后如下,请问这个地方分层是不是不对
我这样的用法对么
see #242
我和你问题一样,怎么解决啊
我用classification的Vamba中的vmamba_base_s1l20作为我的显著目标检测的backbone,在load目标检测与分类权重的时候会提示KeyError: 'layers.0.blocks.0.op.in_proj.weight',并且我在backbone的时候发现,后面分片是有问题的,以vmamba_base_s1l20为例layer传入了四层,四层的通道数大小分别是256, 512, 1024, 1024,是否128作为layer的第一层,就是第1480行的x = self.patch_embed(x)这个代码,不知道我理解的是否有问题