How to use coarsely annotated data for cityscapes

HRNet / HRNet-Semantic-Segmentation

The OCR approach is rephrased as Segmentation Transformer: https://arxiv.org/abs/1909.11065. This is an official implementation of semantic segmentation for HRNet. https://arxiv.org/abs/1908.07919

Other

3.15k stars 690 forks source link

How to use coarsely annotated data for cityscapes #93

Open verymadmatt opened 4 years ago

verymadmatt commented 4 years ago

Just wondering how the coarsely annotated cityscapes data is used for training. Directly putting them into "train list" will make the training time 6.7 times longer since the coarsely annotated dataset is 6.7 times larger. It's almost unbearable.

Besides, not sure if the following parameters in the config file are related with the extra training data? Could they be used for coarsely annotated cityscapes data? EXTRA_TRAIN_SET, EXTRA_EPOCH, EXTRA_LR

Any advice will be much appreciated.

doulemint commented 3 years ago

Hi @verymadmatt, did you figure out how to train coarsely annotated cityscapes data? I also met this question.