The model can be finetuned for other tasks such as panoptic segmentation with relatively little effort, comparing favorably to the best methods.

Cellpose-SAM can be finetuned or retrained with the human-in-the-loop approach from Cellpose 2, which can be useful both in 2D and 3D.

The algorithm is invariant to channel order and cell size, and is robust to many common image degradations such as shot noise, (an)isotropic blur and under-sampling.

Cellpose-SAM outperforms human-to-human agreement and approaches an estimated absolute bound on performance.

We combined:
1) SAM, a pretrained foundation model with good generalization properties.
2) Cellpose, a good cellular segmentation framework.

🚀🔬🦠 Releasing 🤖Cellpose-SAM🤖, a cellular segmentation algorithm with superhuman generalization 🦸‍♀️. Try it now on 🤗 huggingface.co/spaces/mouse...

paper: www.biorxiv.org/content/10.1...
w/ @computingnature.bsky.social 1/n