The shape of pred does not match that of gt

[CBHCK]

Hello! What a wonderful job! However, I encountered some problems during the running process. I executed train_net through the -eval-only method. But in the sem_seg_evaluation.py file of detectron2, I noticed that the shape of pred did not match the shape of gt. I checked and found that the annotation released by isaid is in rgb png format, while the output of pred has no rgb dimension. How did you handle this issue? I'm very sorry that I couldn't find a solution in the code. Hope to get a reply!

[zqcrafts]

Thank you for your attention. You may need to check whether the isaid needs to be converted to a single channel. In addition, our training code and dataset have also been open sourced. You can refer to the training on the sky_sa dataset and try debugging.

[CBHCK]

感谢您的关注。您可能需要检查 isaid 是否需要转换为单通道。此外，我们的训练代码和数据集也已开源。您可以参考 sky_sa 数据集上的训练并尝试调试。

THK！