Retention-Constrained Post-Training Quantization of Cellpose-SAM for Stem Cell Microscopy
Abstract
Induced pluripotent stem cell (iPSC) culture increasingly relies on segmentation foundation models, yet deployment on laboratory CPUs and edge hardware requires compression schemes that are both efficient and auditable. We present a deployment-oriented evaluation of compressed Cellpose-SAM using a pre-specified retention criterion: the 95% cluster-bootstrap interval of mean change from FP32 must remain 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, weight-only W8A16 preserves instance F1 across all modalities. A sensitivity-guided mixed W4/W8 scheme, using four INT8 exceptions, achieves a 6.76x reduction in weight storage with no observed catastrophic failures (0/176 fields), matching W8A16 at this sample size. In contrast, ternary weight-only quantization achieves 12.08x compression but fails catastrophically on 169/176 fields. These results demonstrate that compression should be evaluated by modality-stratified downstream retention rather than single-number accuracy, and establish a reproducible protocol for auditing compressed foundation models in regulated stem-cell imaging.
Related
- Co-retained intron retention on single RNA molecules marks fitness-critical genes
- Atlas of adult Danionella fish telencephalon reveals retention of larval characters
- Removal of Off-Season Inflorescences Improves Fruit Retention, Kernel Quality, and Source–Sink Balance in Macadamia ‘O.C’
- Contrasting Green Manure Species and Return Methods Regulate Soil Organic Carbon Mineralization and Retention in Coastal Saline Soil
- The role of sleep in the retention and long-term maintenance of newly formed ability self-beliefs
BibTeX
@misc{romero2026retentionconstrained,
title = {{Retention-Constrained Post-Training Quantization of Cellpose-SAM for Stem Cell Microscopy}},
author = {Sebastián A. Cruz Romero},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.21038},
eprint = {2609.21038},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.21038},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Retention-Constrained Post-Training Quantization of Cellpose-SAM for Stem Cell Microscopy
AU - Sebastián A. Cruz Romero
PY - 2026
DA - 2026/09/17
PB - arXiv
DO - 10.48550/arxiv.2609.21038
UR - https://arxiv.org/abs/2609.21038
AB - Induced pluripotent stem cell (iPSC) culture increasingly relies on segmentation foundation models, yet deployment on laboratory CPUs and edge hardware requires compression schemes that are both efficient and auditable. We present a deployment-oriented evaluation of compressed Cellpose-SAM using a pre-specified retention criterion: the 95% cluster-bootstrap interval of mean change from FP32 must remain 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, weight-only W8A16 preserves instance F1 across all modalities. A sensitivity-guided mixed W4/W8 scheme, using four INT8 exceptions, achieves a 6.76x reduction in weight storage with no observed catastrophic failures (0/176 fields), matching W8A16 at this sample size. In contrast, ternary weight-only quantization achieves 12.08x compression but fails catastrophically on 169/176 fields. These results demonstrate that compression should be evaluated by modality-stratified downstream retention rather than single-number accuracy, and establish a reproducible protocol for auditing compressed foundation models in regulated stem-cell imaging.
ER -