search

charlesCXK / TorchSemiSeg

[CVPR 2021] CPS: Semi-Supervised Semantic Segmentation with Cross Pseudo Supervision
MIT License
501 stars 73 forks source link

overlap in you supervised and unsupervised data-set, for few supervision data #20

Closed ghost closed 3 years ago

ghost commented 3 years ago

In you few-supervise setting, which is in here which seems is labeled data but over lap with here.

Also, in you 1/8 ratio, labeled with unlabeled

in you 1/4 ratio, labeled with unlabeled,

in you 1/2 ratio, https://github.com/charlesCXK/TorchSemiSeg/blob/bab3a31ef7a5b59b932820b4d5d93a1b8bfc41ec/DATA/pascal_voc/subset_train_aug/pseudoseg_labeled_1-2.txt#L732 and https://github.com/charlesCXK/TorchSemiSeg/blob/bab3a31ef7a5b59b932820b4d5d93a1b8bfc41ec/DATA/pascal_voc/subset_train_aug/pseudoseg_unlabeled_1-2.txt#L321.

In each ratio, there is one overlap, and I suspect it should be a mis-operation.

ghost commented 3 years ago

Sorry, it shouldn't be a big deal. As the overlapped data is counted as supervised part. It just confused me a bit, all good.

charlesCXK commented 3 years ago

@Sun-Rider Hi, I understand what you are worrying about. However, the unlabeled id files (e.g., pseudoseg_unlabeled_1-2.txt) are not used in our experiments. In Tab. 5, we follow PseudoSeg and view the whole augmented set of VOC as the unlabeled set.

ghost commented 3 years ago

@charlesCXK Thanks so much for the clarification, that makes more sense.

charlesCXK commented 3 years ago

Hi, the config file of few supervision could be found here. Hope this would help you.