Unified Image Processing Toolkit

One C++ toolkit for enhancement, restoration, detection, segmentation, tracking, and style workflows on images and videos.

Built with ONNX Runtime, OpenCV, and FFmpeg, with a lightweight local web UI for quick testing and deployment.

Demo

Super Resolution	Dehaze

Derain	Underwater

Colorization	Style Transfer

Tracking

AI_Result_track_yolov8n_track.mp4

What It Does

• Image and video processing in one local C++ application
• ONNX Runtime inference backend
• OpenCV-based preprocessing, rendering, and postprocessing
• YOLO-based detection, segmentation, and tracking
• Restoration workflows for super resolution, dehaze, derain, desnow, underwater enhancement, and old-photo repair
• Colorization and style transfer support
• Lightweight browser UI served directly by the C++ backend

Supported Tasks

superres · dehaze · derain · desnow · underwater · old_photo · colorization · style · detect · segment · track

Quick Start

Requirements

• C++17 compiler
• CMake 3.16+
• ONNX Runtime
• OpenCV
• FFmpeg

Build

cmake -S . -B build
cmake --build build --config Release

Run

Start the executable and open the local UI in your browser:

http://localhost:8080

Downloads

• ONNX Runtime
• OpenCV
• FFmpeg
• Weights pack

Project Structure

.
+-- CMakeLists.txt
+-- README.md
+-- LICENSE
+-- include/
|   +-- basemodel.h
|   +-- colorizer.h
|   +-- draw_utils.h
|   +-- httplib.h
|   +-- ModelTypes.h
|   +-- restoration.h
|   +-- tracker.h
|   \-- yolo.h
+-- src/
|   +-- basemodel.cpp
|   +-- colorizer.cpp
|   +-- main.cpp
|   +-- restoration.cpp
|   \-- yolo.cpp
+-- web/
|   \-- index.html
+-- figs/
\-- weights/   # local only, not committed

Notes

• Model weights are not committed to this repository.
• Third-party runtime binaries should be downloaded separately.

License

Apache License 2.0.