sr100_person_segmentation_480x640

Person Segmentation 480x640 (SR100 Series)

Model Overview

The Person Segmentation 480x640 model developed by Synaptics, is a lightweight quantized tflite model developed for the SR100 processor in the Synaptics Astra™ SR MCU Series.

The output includes segmented regions that represent the exact shape of each person in the image, providing both segmentation and confidence-level insights for each detection.

Model Features

• Model Type: Person Segmentation
• Input Size: 480x640
• Output: For each detected person, a segmentation mask outlining along with confidence scores for each region.
• Classes: Single class (person); specifically designed for person segmentation.

Deployment on Astra

You can optimize this model for Synaptics Astra SR100 MCU using our our hosted SR100 Model Compiler HF Space.

• Processor: Astra™ SR100 MCU
• Platform: Astra™ Machina Micro Dev Kit
• Quantization: INT8 (fully quantized)
• Compiler: SR100 Model Compiler
• Preprocessing: Input images must be resized and quantized to match model requirements

Intended Applications

This model enables real-time person segmentation for embedded edge devices. Example use cases include:

• Human presence and activity monitoring
• Fitness and wellness tracking
• Smart home automation
• Interactive user interfaces

Evaluate Model

You can evaluate and test this model directly in our hosted Hugging Face Space, optimized for Synaptics SR110 MCU. This space provides a seamless sandbox for model evaluation using hardware-specific quantization and runtime settings.

For a detailed walkthrough on how to optimize and evaluate a model, please see our Evaluate Model Guide page.

To get started quickly with Astra SR Series, visit our SR Quick Start page.

License

Distributed under the Apache License 2.0, allowing flexible use, modification, and distribution.

Learn More

• Synaptics AI Developer Zone: Get started with documentation, tutorials and resources for your Edge AI journey.
• Astra Support Portal: Connect with our engineering team and community.