rom1504 / clip-retrieval

Easily compute clip embeddings and build a clip retrieval system with them
https://rom1504.github.io/clip-retrieval/
MIT License
2.42k stars 213 forks source link

About image2image retrieval on custom datasets #265

Closed zbw4034 closed 1 year ago

zbw4034 commented 1 year ago

Hi! Thanks for your fantastic and helpful work! It's apprecated by everyone in my lab.

I plan to conduct image-to-image retrieval on my own local dataset(like Visual Genome and MSCOCO), and use knn to return several similar images. I find this function is already supported on your clip front API with webdata. I read the readme and it seems it's mainly about image-text retrieval. Could u tell me how to implement?

rom1504 commented 1 year ago

Hi,

Glad it's helpful

The clip back code allows computing image embeddings from images the user provides and then using that as a query for the knn index. That's image image search

You can use clip inference with only images and it will do what you want

On Mon, May 8, 2023, 07:39 zbw4034 @.***> wrote:

Hi! Thanks for your fantastic and helpful work! It's apprecated by everyone in my lab.

I plan to conduct image-to-image retrieval on my own local dataset(like Visual Genome and MSCOCO), and use knn to return several similar images. I find this function is already supported on your clip front API with webdata. I read the readme and it seems it's mainly about image-text retrieval. Could u tell me how to implement?

— Reply to this email directly, view it on GitHub https://github.com/rom1504/clip-retrieval/issues/265, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAR437SL5P5QG55WUXPRVCTXFCBK5ANCNFSM6AAAAAAXZOW6OI . You are receiving this because you are subscribed to this thread.Message ID: @.***>