MengzhangLi

> When you use a deeper and larger backbone (e.g. R101), things are different and maybe the Spatial Path is necessary to provide detailed spatial information as it has small...

Hi, related issue in MMSegmentation: https://github.com/open-mmlab/mmsegmentation/issues/1068. Our user says it would not fix bug in MMSegmentation but he or she gave the alternative solution. Best,

https://github.com/open-mmlab/mmsegmentation/blob/master/mmseg/models/backbones/beit.py#L21-L24 It needs scipy package.

1. Check other config files which has auxiliary heads settings such as [this one](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/_base_/models/bisenetv1_r18-d32.py#L40-L65). Use `in_index` to determine which feature map from backbone as input for certain auxiliary head. You...

Seems like the path is wrong, Win OS and Linux OS is different. `'D:/SCM/yg/is\\test\\images\\part1.zip` shows two OS path. Try to set `python tools/convert_datasets.py/isaid.py D:\SCM\yg\is`.

Please use English or English & Chinese for wider discussion in community.

> Please read the [data preparation](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/dataset_prepare.md#prepare-datasets) first. 

(1) The error is caused by unmatched parameters in model and PTH file, please check the difference with your model and `fcn_unet_s5-d16_ce-1.0-dice-3.0_64x64_40k_drive`. (2) It depends on your dataset, `fcn_unet_s5-d16_ce-1.0-dice-3.0_64x64_40k_drive` is...

For single image, you could use command [here](https://github.com/open-mmlab/mmsegmentation/blob/master/demo/inference_demo.ipynb). For your customized dataset, because our demo just only supports guidance for beginners so it mainly supports single image rather than whole...

Sorry for late reply. Just use `best_mIoU_iter_200.pth` as checkpoint and run `python tools/test.py ${CONFIG} ${CHECKPOINT} --eval mIoU`.