search

raoyongming / DenseCLIP

[CVPR 2022] DenseCLIP: Language-Guided Dense Prediction with Context-Aware Prompting
505 stars 38 forks source link

About prompt text #31

Closed good-demo closed 1 year ago

good-demo commented 1 year ago

Thanks for your sharing work

I want to know how the prompt text create?(script or manual )

Can you provide some examples？

raoyongming commented 1 year ago

Hi, thanks for your interest in our work. The prompt text is created using the name of semantic classes from the dataset. For example, in the segmentation task, we use the names provided in mmseg to create the class embeddings

good-demo commented 1 year ago

Thanks for your reply

anurag-198 commented 1 year ago

Hi, thanks for your interest in our work. The prompt text is created using the name of semantic classes from the dataset. For example, in the segmentation task, we use the names provided in mmseg to create the class embeddings

Hi, Thanks for your work. In the paper, it is mentioned that you use CoOp to generate prompts where <p1, p2,...,pn> are learnable. But from the code it seems that only the class labels are used for generating the prompt text ? Can you please confirm this, if I am missing something ?