Miao Zheng

Would you mind providing the dataset configuration?

Many thanks for your contribution, please sign [CLA](https://cla-assistant.io/open-mmlab/mmsegmentation?pullRequest=1849) first.

You might try to change `num_classes` in model cfg ref: https://github.com/open-mmlab/mmsegmentation/blob/4eaa8e69191cc293b64dafe47f1f88a7d468c93c/configs/_base_/models/deeplabv3_r50-d8.py#L24

Might give us more clear comparison? like what are the `'other models'`?

Would you mind providing the config file and running stript you used?

Could you provide the error log? We can know how image size mismatch.

According to the error `TypeError: "img" must be a numpy array or a str or a pathlib.Path object` Please check the data type of input image

Could you please share with us which part of the document you referred?

I prefer approach 1, as the `RandomFlip` api is much easier to use. #1812 might increase complexity.

From the error log, I guess you trained the model with your own datasets, and the seg map label shape is (640,640,3) I suggest you might add a suggest break...