# PicoSAM2: Low-Latency Segmentation In-Sensor for Edge Vision Applications

arXiv: https://arxiv.org/abs/2506.18807 · HTML: https://mlsanity.com/paper/2506.18807/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 40 references that are in the map.
- First posted on arXiv 23 Jun 2025.

## Builds on (most cited first, 22 of 22)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [Xception: Deep Learning with Depthwise Separable Convolutions](https://mlsanity.com/paper/1610.02357/) (2016)
- [V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation](https://mlsanity.com/paper/1606.04797/) (2016)
- [LVIS: A Dataset for Large Vocabulary Instance Segmentation](https://mlsanity.com/paper/1908.03195/) (2019)
- [A Survey of Quantization Methods for Efficient Neural Network Inference](https://mlsanity.com/paper/2103.13630/) (2021)
- [Up or Down? Adaptive Rounding for Post-Training Quantization](https://mlsanity.com/paper/2004.10568/) (2020)
- [Hiera: A Hierarchical Vision Transformer without the Bells-and-Whistles](https://mlsanity.com/paper/2306.00989/) (2023)
- [Fast Segment Anything](https://mlsanity.com/paper/2306.12156/) (2023)
- [MCUNet: Tiny Deep Learning on IoT Devices](https://mlsanity.com/paper/2007.10319/) (2020)
- [MobileSAMv2: Faster Segment Anything to Everything](https://mlsanity.com/paper/2312.09579/) (2023)
- [TinySAM: Pushing the Envelope for Efficient Segment Anything Model](https://mlsanity.com/paper/2312.13789/) (2023)
- [FANN-on-MCU: An Open-Source Toolkit for Energy-Efficient Neural Network Inference at the Edge of the Internet of Things](https://mlsanity.com/paper/1911.03314/) (2019)
- [TinyTracker: Ultra-Fast and Ultra-Low-Power Edge Vision In-Sensor for Gaze Estimation](https://mlsanity.com/paper/2307.07813/) (2023)
- [Ultra-Efficient On-Device Object Detection on AI-Integrated Smart Glasses with TinyissimoYOLO](https://mlsanity.com/paper/2311.01057/) (2023)
- [HPTQ: Hardware-Friendly Post Training Quantization](https://mlsanity.com/paper/2109.09113/) (2021)
- [EPTQ: Enhanced Post-Training Quantization via Hessian-guided Network-wise Optimization](https://mlsanity.com/paper/2309.11531/) (2023)
- [Lite-SAM Is Actually What You Need for Segment Everything](https://mlsanity.com/paper/2407.08965/) (2024)
- [Data Generation for Hardware-Friendly Post-Training Quantization](https://mlsanity.com/paper/2410.22110/) (2024)

## Built on by (most cited first, 6 of 6)

- [Performance Analysis of Edge and In-Sensor AI Processors: A Comparative Review](https://mlsanity.com/paper/2603.08725/) (2026)
- [VCA: Vision-Click-Action Framework for Precise Manipulation of Segmented Objects in Target Ambiguous Environments](https://mlsanity.com/paper/2602.23583/) (2026)
- [TinyGLASS: Real-Time Self-Supervised In-Sensor Anomaly Detection](https://mlsanity.com/paper/2603.16451/) (2026)
- [PicoSAM3: Real-Time In-Sensor Region-of-Interest Segmentation](https://mlsanity.com/paper/2603.11917/) (2026)
- [Promptable Fire Segmentation: Unleashing SAM2's Potential for Real-Time Mobile Deployment with Strategic Bounding Box Guidance](https://mlsanity.com/paper/2510.21782/) (2025)
- [Exploiting In-Sensor Computing for Energy-Efficient Earth Observation](https://mlsanity.com/paper/2606.01271/) (2026)

## Abstract (arXiv)

Real-time, on-device segmentation is critical for latency-sensitive and privacy-aware applications like smart glasses and IoT devices. We introduce PicoSAM2, a lightweight (1.3M parameters, 336M MACs) promptable segmentation model optimized for edge and in-sensor execution, including the Sony IMX500. It builds on a depthwise separable U-Net, with knowledge distillation and fixed-point prompt encoding to learn from the Segment Anything Model 2 (SAM2). On COCO and LVIS, it achieves 51.9% and 44.9% mIoU, respectively. The quantized model (1.22MB) runs at 14.3 ms on the IMX500-achieving 86 MACs/cycle, making it the only model meeting both memory and compute constraints for in-sensor deployment. Distillation boosts LVIS performance by +3.5% mIoU and +5.1% mAP. These results demonstrate that efficient, promptable segmentation is feasible directly on-camera, enabling privacy-preserving vision without cloud or host processing.