ofxCoral

OpenFrameworks addon for Google Coral Edge TPU inference using TensorFlow Lite.

Features

• Edge TPU acceleration - Automatic detection and use of Google Coral USB Accelerator / Dev Board
• Camera integration - GStreamer-based camera capture with v4l2src
• Classification & Detection - Support for both image classification and object detection models
• Cross-platform - Works on Linux (x64, ARMv7, ARM64), with stubs for other platforms

Requirements

• openFrameworks 0.12+
• Google Coral Edge TPU runtime (libedgetpu1, libedgetpu-dev)
• TensorFlow Lite (libtensorflow-lite)
• GStreamer 1.0 with v4l2 and appsink plugins
• Compatible Coral model (.tflite compiled for Edge TPU)

Install dependencies (Raspberry Pi / Debian arm64)

# Add Coral repository
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt-get update

Install Edge TPU runtime and TensorFlow Lite
sudo apt-get install libedgetpu1-std libedgetpu-dev
sudo apt-get install python3-tflite-runtime  # includes libtensorflow-lite

GStreamer
sudo apt-get install libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev \
    gstreamer1.0-plugins-base gstreamer1.0-plugins-good \
    gstreamer1.0-plugins-bad gstreamer1.0-libav

Download models

cd /home/pi/openFrameworks/of_v0.12.1_linuxaarch64_release/addons/ofxCoral
sh docs/download_models.sh
Models will be in docs/all_models/

Usage

Basic setup

#include "ofxCoral.h"

ofxCoral::ofxCoral coral;

void setup() {
    // Load model and labels
    coral.setup("path/to/model_edgetpu.tflite", "path/to/labels.txt", true);

    // Setup camera (optional - can also run inference on static images)
    coral.setupCamera(640, 480, 30, "/dev/video0");

    // Set callbacks
    coral.onClassification = [](const std::vector<ofxCoral::Classification>& results) {
        for (const auto& c : results) {
            ofLogNotice() << c.label << ": " << c.confidence;
        }
    };

    coral.onDetection = [](const std::vector<ofxCoral::Detection>& results) {
        for (const auto& d : results) {
            ofLogNotice() << d.label << ": " << d.confidence
                          << " @" << d.x << "," << d.y << " " << d.width << "x" << d.height;
        }
    };

    // Start the pipeline
    coral.start();
}

void update() {
    coral.update();
}

void draw() {
    // Draw results...
}

void exit() {
    coral.stop();
}

Manual inference on images

ofPixels pixels;
// ... load or capture pixels ...

// Classification
auto results = coral.classify(pixels);
for (const auto& c : results) {
    cout << c.label << ": " << c.confidence << endl;
}

// Detection (if model supports it)
auto detections = coral.detect(pixels);
for (const auto& d : detections) {
    cout << d.label << ": " << d.confidence
         << " bbox: " << d.x << "," << d.y << " " << d.width << "x" << d.height << endl;
}

Model Compatibility

Classification models (ImageNet, iNaturalist)

• Input: 224x224x3 RGB
• Output: Single tensor with class probabilities
• Examples: MobileNet v1/v2, Inception v1-v4

Detection models (SSD)

• Input: 300x300x3 RGB (varies by model)
• Output: 4 tensors [num_detections, boxes, classes, scores]
• Examples: MobileNet SSD v1/v2 COCO, MobileNet SSD v2 Face

Architecture

ofxCoral (main class)
├── InferenceEngine - TensorFlow Lite + Edge TPU delegate
│   ├── Model loading (.tflite)
│   ├── Label loading
│   ├── Edge TPU delegate attachment
│   ├── classify() - returns top-K classifications
│   └── detect() - returns bounding box detections
└── CameraStreamer - GStreamer pipeline
    ├── v4l2src camera capture
    ├── videoconvert to RGB
    ├── appsink callback to inference
    └── Runs in separate thread

Platform Notes

Platform	Edge TPU	Camera	Status
Linux x64	❌	✅	CPU only
Linux ARMv7	✅	✅	Full support
Linux ARM64	✅	✅	Full support (Pi 4, Coral Dev Board)
Android	✅	✅	Requires NDK libs
iOS	❌	❌	Not supported
Emscripten	❌	❌	Not supported

*Android requires prebuilt libedgetpu.so and libtensorflow-lite.so in libs/<arch>/

License

MIT - See LICENSE file

References

• Coral Examples
• TensorFlow Lite C++ API
• Edge TPU Python API