Skip to content

README.md

Latest commit

 

History

History
80 lines (54 loc) · 1.97 KB

README.md

File metadata and controls

80 lines (54 loc) · 1.97 KB

🖥️ AnomaVision C++ Inference

This folder contains the C++ implementation of AnomaVision inference using ONNX Runtime and OpenCV. It is designed for real-time anomaly detection on edge devices and environments where Python is not available.

📦 Requirements

⚙️ Build

# From project root
cmake -S . -B build
cmake --build build --config Release

Make sure to update CMakeLists.txt with the correct paths to ONNX Runtime and OpenCV on your system.

▶️ Run Inference

./build/Release/onnx_inference.exe model.onnx D:/01-DATA/test \
    --save_viz D:/output --alpha 0.5 --thresh 13.0

Options:

  • --save_viz out_dir → save annotated results
  • --alpha → heatmap blending factor (default: 0.5)
  • --thresh → anomaly threshold (default: 13.0)

📊 Example Output

Found 10 images

Processing 000.png (1/10)
Score: 29.88  Anomalous: YES  Time: 31.53 ms
Number of anomalous pixels: 4271  Ratio: 0.085

Processing 007.png (2/10)
Score: 59.71  Anomalous: YES  Time: 33.93 ms
Number of anomalous pixels: 6068  Ratio: 0.121
...
  • Average inference time ≈ 38.9 ms (~25.7 FPS)
  • Visualization windows display heatmaps + bounding boxes
  • Results saved if --save_viz is provided

🧩 Code Structure

  • detect.cpp → entry point (CLI parser, launches app)

  • oop_anomaly_detector.h/.cpp → modular pipeline

    • Config → runtime parameters
    • Preprocessor → image normalization & resize
    • ONNXModel → inference with ONNX Runtime
    • Postprocessor → thresholding, scores, masks
    • Visualizer → overlay, annotations, bounding boxes
    • App → orchestrates the full workflow

✅ This C++ module makes AnomaVision truly edge-ready, combining Python flexibility with C++ performance.