[Request] Benchmark across different supported models

[meet-minimalist]

Hi, I saw that you support multiple models to get the embeddings. https://visual-layer.readme.io/docs/using-your-own-model#running-fastdup-with-a-preconfigured--model

Do you have any benchmarks or any reference for papers which compares these models?

Any guide on when to use what type of model. This would help a lot.

[dnth]

Hi @meet-minimalist as far as I know the best model varies from one case to another. For example, the dinov2 model might perform well on natural images, but maybe not on some niche datasets like medical X-ray dataset.

For niche domains, you may find that the models used to train on the dataset may perform better than general off the shelves model. We don't have a benchmark comparison currently as the result may vary from one domain to another.