Retention-Constrained Post-Training Quantization of Cellpose-SAM for Stem Cell Microscopy

• 类型：arxiv
• 标识：2609.21038
• 链接：https://arxiv.org/abs/2609.21038
• 主分类：engineering
• 形态：benchmark
• TLDR：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-onl
• 副分类：llm-infra
• 待LLM分类：否
• 标题中文：干细胞显微成像中 Cellpose-SAM 的保留约束训练后量化
• TLDR中文：诱导多能干细胞 (iPSC) 培养日益依赖分割基础模型，但在实验室 CPU 和边缘硬件上的部署需要既高效又可审计的压缩方案。我们对压缩后的 Cellpose-SAM 进行了面向部署的评估，采用预指定的保留标准：相对 FP32 均值变化的 95% 聚类自助区间在每种成像模态下均须高于固定的 -0.02 边界。在涵盖 BBBC038 细胞核、BBBC039 U2OS 荧光和 NIST iPSC 图像、覆盖不同密度区间的分层 176 视野面板上，仅权重量化
• 来源文件：
• /inbox/tom/_candidates/2026-09-21-agent-rag-longcontext-candidates.json