Hardware Guide

ESP32-S3 for Object Detection with Edge Impulse

Edge Impulse provides an end-to-end pipeline for deploying object detection on the ESP32-S3. You collect images, train a FOMO or MobileNet-SSD model in the cloud, and export an optimized C++ library that runs inference at 5-10 FPS on the S3's dual-core Xtensa LX7.

Hardware Specs

Spec ESP32-S3
Processor Dual-core Xtensa LX7 @ 240 MHz
SRAM 512 KB
Flash Up to 16 MB (external)
Key Features Vector instructions (SIMD), USB OTG, LCD/Camera interface, Up to 8 MB PSRAM
Connectivity Wi-Fi 802.11 b/g/n, Bluetooth 5.0 LE
Price Range $3 - $8 (chip), $10 - $25 (dev board)

Compatibility: Excellent

Edge Impulse has first-class ESP32-S3 support with a dedicated firmware target. The FOMO (Faster Objects, More Objects) architecture was specifically designed for constrained devices and runs faster than standard MobileNet-SSD on the S3. With 512 KB SRAM and vector instructions, the ESP32-S3 handles Edge Impulse's exported SDK comfortably. The SDK binary is larger than raw TFLite Micro (~100 KB overhead for the full inference pipeline), but 16 MB external flash eliminates storage concerns. Edge Impulse's data collection tools work directly with the ESP32-S3-EYE's built-in camera, closing the loop from data capture to deployment without external tools.

Getting Started

  1. 1

    Create an Edge Impulse project

    Sign up at edgeimpulse.com and create a new image classification or object detection project. Select ESP32-S3 as your target device in project settings.

  2. 2

    Collect and label training data

    Use the Edge Impulse CLI (edge-impulse-daemon) to stream images directly from an ESP32-S3-EYE board, or upload images from your dataset. Label bounding boxes in the web UI.

  3. 3

    Train and optimize the model

    Choose FOMO for speed or MobileNet-SSD for accuracy. Edge Impulse automatically quantizes to int8 and optimizes for the Xtensa architecture. Check the estimated RAM and latency in the deployment preview.

  4. 4

    Deploy to ESP32-S3

    Export as ESP-IDF library from the Deployment tab. Add the generated SDK to your project's components/ directory. The library includes the model, preprocessing, and inference loop — call run_classifier() from your application code.

Alternatives

ESP32-S3 with TFLite Micro

More control over the model pipeline and smaller binary size, but requires manual model training, quantization, and deployment. Better for teams with ML experience.

STM32H7 with TFLite Micro

Double the RAM (1 MB) and faster clock (480 MHz) enable higher-resolution detection. No Wi-Fi, higher cost, but better raw inference performance.

Explore More

More ESP32-S3 guides More Object Detection guides All resources Find the right MCU

FAQ

Is Edge Impulse free for ESP32-S3 object detection projects?
Edge Impulse is free for individual developers with up to 5 projects. The free tier includes model training, optimization, and ESP32-S3 deployment. Enterprise features like team collaboration and custom deployment targets require a paid plan.
What is FOMO and how does it compare to MobileNet-SSD on ESP32-S3?
FOMO (Faster Objects, More Objects) is Edge Impulse's lightweight object detection architecture. On the ESP32-S3, FOMO runs at 10+ FPS versus 2-5 FPS for MobileNet-SSD. The tradeoff is that FOMO provides object centroids instead of full bounding boxes.
Can I collect training data directly from the ESP32-S3?
Yes. The Edge Impulse CLI daemon connects to the ESP32-S3-EYE's camera over serial and streams images directly into your project. This ensures training data matches the deployment camera's characteristics.

Deploy Edge Impulse Models with ForestHub

Orchestrate your Edge Impulse detection pipeline visually — from sensor input to action output on the ESP32-S3.

Get Started Free