BioCLIP 2 image encoder, converted for phones

The image encoder from BioCLIP 2 by Imageomics, converted for the Thicket bird identification app, which identifies birds from photos on the phone.

BioCLIP 2 identifies a photo by comparing its image embedding with a text embedding for each species. Only the image encoder is here; the app carries its own text embeddings, one per species it knows. Each model takes a 224 × 224 RGB image, normalised with BioCLIP 2's mean and standard deviation, and returns a unit-length 768-dimensional embedding.

File For Weights SHA-256
bioclip2_image.mlpackage.zip iOS (Core ML), a zipped .mlpackage 8-bit 11f5f34e0b533c0d8cac74772d3a9f77a894bbdb461f2300201289c4127a56b1
bioclip2_image.tflite Android (LiteRT / TensorFlow Lite) 8-bit 97afbcd418a754621ae40314de288d815249e8ce6d01e11869eade5f1b1c53b1

On 116 labelled bird photos, cropped to the bird, the 8-bit Core ML model put the right species first 91.4% of the time and in the top five 97.4% of the time, against 88.8% and 97.4% for the unconverted model.

Licence

MIT, as BioCLIP 2 is. The model is Imageomics' work; only the conversion is new. If you use it, cite BioCLIP 2 as its model card asks, both the model and the paper:

@software{Gu_BioCLIP_2_model,
  author = {Jianyang Gu and Samuel Stevens and Elizabeth G Campolongo and Matthew J Thompson and Net Zhang and Jiaman Wu and Andrei Kopanev and Zheda Mai and Alexander E. White and James Balhoff and Wasila M Dahdul and Daniel Rubenstein and Hilmar Lapp and Tanya Berger-Wolf and Wei-Lun Chao and Yu Su},
  license = {MIT},
  title = {{BioCLIP 2}},
  url = {https://huggingface.co/imageomics/bioclip-2},
  version = {1.0.0},
  doi = {10.57967/hf/5765},
  publisher = {Hugging Face},
  year = {2025}
}

@inproceedings{NEURIPS2025_94da80cb,
 author = {Gu, Jianyang and Stevens, Sam and Campolongo, Elizabeth and Thompson, Matthew and Zhang, Net and Wu, Jiaman and Kopanev, Andrei and Mai, Zheda and White, Alexander and Balhoff, James and Dahdul, Wasila and Rubenstein, Daniel and Lapp, Hilmar and Berger-Wolf, Tanya and Chao, Wei-Lun (Harry) and Su, Yu},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {D. Belgrave and C. Zhang and H. Lin and R. Pascanu and P. Koniusz and M. Ghassemi and N. Chen},
 pages = {102778--102811},
 publisher = {Curran Associates, Inc.},
 title = {BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning},
 url = {https://proceedings.neurips.cc/paper_files/paper/2025/file/94da80cbfe870c1db958c88a8a27018c-Paper-Conference.pdf},
 volume = {38},
 year = {2025}
}
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for fearnt/thicket-bioclip2

Finetuned
(4)
this model