search

ZYN-1101 / DandR

MIT License
10 stars 5 forks source link

text embedding generation #4

Open xiaoyangbenjiance opened 7 months ago

xiaoyangbenjiance commented 7 months ago

can you provide the text embedding generation .

SijieLuo commented 5 days ago

can you provide the text embedding generation .

ZYN-1101 commented 5 days ago

Refer to the official implementation of CLIP for the text embedding generation. For the foreground classes, I recommend using "a photo of a [classname]" as a prompt to generate the category embedding. For the background category, we can directly input "background" into the text encoder of CLIP to get the text embedding.