NERC-CEH / plankton_ml

A project for image processing and analysis pipelines for plankton sampling
GNU General Public License v3.0
0 stars 1 forks source link

Extract and store embeddings from BioCLIP #10

Open metazool opened 4 months ago

metazool commented 4 months ago

BioCLIP is a foundation model designed for classification of all or parts of biological images of all kinds, it could also be a useful source of embeddings that would generalise to a few projects

Aiming at running BioCLIP for the same outcome as #5 could be a great exercise for our regular pair programming session @albags ?

metazool commented 4 months ago

Searching an aerial photo with text queries – a demo and how it works - lovely blog writeup about the potential inherent in this approach, in this case doing text search from embeddings with a model called SkyCLIP that's fine-tuned on remote sensing imagery

metazool commented 2 months ago

https://imageomics.github.io/bioclip/ - BioCLIP paper

metazool commented 2 months ago

https://github.com/Imageomics/bioclip/blob/main/docs/imageomics/BioCLIP_ModelCard.md - it has a model card, which is really good to see - something the Turing Inst projects should be consistently advocating for!

See also BioCLIP on HuggingFace https://huggingface.co/imageomics/bioclip