Miao Zheng

This seems to be a problem with your dataset.

Closing the issue, as there is no activity for a while. We hope your issue has been resolved. If not, please feel free to open a new one.

Would you like to let me know what you want to modify? Network architecture or the output of decode head? If it is architecture, you can modify the `__init__` to...

of course, we have supported very powerful visualization tool in mmseg 1.0.0rc0, you can try it reference https://github.com/open-mmlab/mmengine/blob/cfb884c180e97454b14b1dc12bac21bd376f82db/mmengine/visualization/visualizer.py#L119-L124 https://github.com/open-mmlab/mmsegmentation/blob/1.x/docs/en/user_guides/visualization.md

Just save any tensor map like below as you want, then load it ant see it with the visualization tool. https://github.com/open-mmlab/mmsegmentation/blob/b51670b61339e5b10c5ab6c277de6b6a387fdff0/mmseg/models/decode_heads/dpt_head.py#L287

Would you mind providing the config file you used?

`'unchanged'` means 'unchannge the data', and would not change the channels of ground truth label, so if the label has 1 channel and `'unchanged'` will load it and return a...

> I suggest you save the ground truth with the `unchanged` flag. It is meaningful as the value is label index, not gray color.

Could you provide more information following the issue template?

Please try to replace `MultiScaleFlipAug` with `Normalization` in test_pipeline https://github.com/open-mmlab/mmsegmentation/blob/dd42fa8d0125632371a41a87c20485494c973535/configs/_base_/datasets/ade20k.py#L21-L32 https://github.com/open-mmlab/mmsegmentation/blob/dd42fa8d0125632371a41a87c20485494c973535/configs/_base_/datasets/ade20k.py#L14