search

sheng-eatamath / PromptCAL

Official Implementation of paper: PromptCAL: Contrastive Affinity Learning via Auxiliary Prompts for Generalized Novel Category Discovery (CVPR'23)
MIT License
39 stars 7 forks source link

Question about SemiPriori #9

Closed pgh2874 closed 8 months ago

pgh2874 commented 8 months ago

Hello author! First of all, thank you for the constructive work in Category discovery. I have a question about the way of applying SemiPriori which adds sample-wise class labels as pairwise constraints to G˜ d. As the dataset includes labeled and unlabeled data, unlabeled data cannot add class labels. How could SemiPriori work?

Thanks in advance.

sheng-eatamath commented 8 months ago

Hi, Sorry for my late reply. First of all, thanks for your interest in our work. After we obtain the diffused graph after affinity propagation, affinities between labeled samples are noisy. Therefore, we convert the label information of labeled data into pairwise affinities and calibrate the predicted positive/negative entries on the diffused graph. Note that this operation only affects the entries of pairs of labeled data. You can see more details in our paper and codes.