RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 82 but got size 81 for tensor number 1 in the list.

[vansin]

在 pipeline 升级了随机旋转变换数据增强，请问在pipline自定义的数据增强后，里面的图片尺寸需要满足什么要求？ 运行 browse_dataset.py 是正常的，报了如下错误

Expected size 82 but got size 81 for tensor number 1 in the list.

  File "/project/mmrotate-1.x/mmdetection/mmdet/models/necks/cspnext_pafpn.py", line 153, in forward
    torch.cat([upsample_feat, feat_low], 1))
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 82 but got size 81 for tensor number 1 in the list.

尝试了以下 retinanet 可以正常使用升级后的随机变换数据增强。应该是cspnext_pafpn中上下采样出了问题。

[vansin]

问题已解决