Martin Liao-WHU comments

Results 5 comments of


                                            Martin Liao-WHU

代码运行的时候导入的config在哪里呢

Please refer to the tutorial of the MMsegmentation codebase.

Preprocessed Camvid and PASCAL dataset

Thanks for your interest. I have uploaded the pre-processed camvid/PASCAL datasets to [onedrive](https://whueducn-my.sharepoint.com/:f:/g/personal/martin_liao_whu_edu_cn/EjklDmgVOitPrhuAwy6h6EkBPkyTvnlCkTN0BdjPIIc6xA?e=1i6D4Z) and [baidu-disk](https://pan.baidu.com/s/1DD1LkEaTFUtabbJtTh_8iw?pwd=tpe4) (passwd: tpe4). I will fix the error in the pre-processing code later.

missing 1 required positional argument: 'gt_semantic_sebound'

If you want to train the Mobile-Seed on the custom dataset, you must first generate the semantic boundary label as supervision. Please refer to [link](https://github.com/WHU-USI3DV/Mobile-Seed/blob/main/README.md#L101)(Line 101) for details.

Question about semantic boundary label

We recommend using the ``demoPreproc_gen_png_label.m`` script for training and the ``demoGenGt.m`` script for evaluation. This is because the semantic boundary labels used during training have been bolded. In the MATLAB...

Question about the Class Cityscapes_boundary in mmseg/datasets/cityscapes.py

Inheriting from the ``CityscapesDataset`` is similar. Both are OK. I surmise that the “processing method” implies data pre-processing. For boundary labels and semantic labels, everything remains consistent.