Using LPIPS metric for image retrieval

[woctezuma]

I understand that the model takes as input two images, by design. I would like to know if there is a smart way to use LPIPS metric for image retrieval, other than computing all the pairwise distances.

For information, my dataset of game banners contains about 30k images. In my previous experiments, I extracted image features once, and could then work with this processed data using standard tools for efficient similarity search based on cosine similarity, Minkowski distance, etc.

Thank your for your attention.

[richzhang]

You could save off all the intermediate features (scaled by the learned linear weights), and then use this embedding for retrieval, but it will be very memory expensive. Finding a low-dimensional embedding, consistent with LPIPS distance, would be useful research for this application.