Closed k8xu closed 2 months ago
Hi. I'm quite busy with a few deadlines right now and I cannot promise the release before the end of June'24.
If you need this right now, it is very easy to compute. First, collect a few paintings from the artist in question. Compute the image embeddings using CSD and average it to get the proto representation. Then generate a few images using a diffusion model with the artist's name in the prompt. Do the same procedure as above and compute the average embedding for these generated images.
Then compute the dot-product similarity between these 2 vectors. Check out Figure 2/section 2 to get a sense of what different similarities might mean.
I came across this repository and found this work really interesting, impressive, and awesome. By the way, I want to promote our works on the style descriptor here:
This repository (https://github.com/WangWenhao0716/AnyPatternStyle) seems provide an easy-to-implement style extractor for artworks.
I came across this repository and found this work really interesting, impressive, and awesome. By the way, I want to promote our works on the style descriptor here:
This repository (https://github.com/WangWenhao0716/AnyPatternStyle) seems provide an easy-to-implement style extractor for artworks.
The result is not satisfied in comparison with CSD. I evaluated on several image set with or without same style.
Thanks, I agree that. My work is designed for Image Copy Detection tasks, and the style descriptor is just a generalized application. It is reasonable that it performs worse than the specialized style descriptor CSD. Thanks for your interest again.
@k8xu Do you still need this, or were you able to solve this? I would like to close this issue
Closing this due to inactivity. Re-open if required.
I'm interested in trying CSD for computing similarity score against any given generation, as mentioned in the pending items. When will this code be released? Thank you!