Quantized Cellpose-SAM: 6.76x Smaller With No Failures on Stem Cell Microscopy

gsarti_ · x · 2026-09-21

capicu-ai presents a deployment-oriented post-training quantization evaluation of Cellpose-SAM, aiming to run the iPSC microscopy segmentation foundation model on lab CPUs and edge hardware, with an auditable compression acceptance criterion.

Setup: a pre-specified retention criterion — the 95% cluster-bootstrap interval of mean change from FP32 must stay above a fixed -0.02 margin for every imaging modality. On a stratified 176-field panel spanning BBBC038 nuclei, BBBC039 U2OS fluorescence and NIST iPSC images across density regimes:

Takeaway: compressed foundation models should be evaluated by modality-stratified downstream retention rather than single-number accuracy; the paper also establishes a reproducible protocol for auditing compressed models in regulated stem-cell imaging.

Related event: Cellpose-SAM Quantization Achieves 6.76x Compression With Zero Failures(2 posts)→

Original post →

More from Research

Research channel →