I have a question about ZegCLIP applying to VOC dataset. From the code, ZegCLIP didn't use the data_convertor for VOC dataset as "python tools/dataset_converters/voc_aug.py data/VOCdevkit" as in webpage: https://mmsegmentation.readthedocs.io/en/latest/user_guides/2_dataset_prepare.html. The original VOC semantic segmentation labels are colorful rather than the class numbers recommended by mmsegmentation. The ZegCLIP code works with the above colorful original VOC labels. My question is which code in ZegCLIP transforms the colorful segmentation to class number segmentation? Could anyone help? Thanks a lot!!
I have a question about ZegCLIP applying to VOC dataset. From the code, ZegCLIP didn't use the data_convertor for VOC dataset as "python tools/dataset_converters/voc_aug.py data/VOCdevkit" as in webpage: https://mmsegmentation.readthedocs.io/en/latest/user_guides/2_dataset_prepare.html. The original VOC semantic segmentation labels are colorful rather than the class numbers recommended by mmsegmentation. The ZegCLIP code works with the above colorful original VOC labels. My question is which code in ZegCLIP transforms the colorful segmentation to class number segmentation? Could anyone help? Thanks a lot!!