Skip to content

README_EN.md

Latest commit

 

History

History
121 lines (91 loc) · 3.97 KB

README_EN.md

File metadata and controls

121 lines (91 loc) · 3.97 KB

[Japanese/English]

onnx-cv-graph

An experimental repository for implementing image processing using ONNX graphs.
Each operation is exported as an ONNX file and included in the repository.

image

Web Demo

An inference demo using the exported ONNX models is available at the following page.
*See below for Python inference demo.

You can also perform image processing using onnx-cv-graph through a visual node editor available in the following repository.
An experimental feature to export the constructed graph as a single ONNX file is also implemented.

Features

  • Image processing implemented entirely using ONNX operators
  • Each operation is exported as an ONNX file and included in the repository
  • Multiple operations can be chained and exported as a single ONNX file

Purpose of This Repository

This repository aims to explore the following:

  • Validating image processing that can be implemented purely with ONNX operations
  • Ensuring cross-platform and cross-language reproducibility via ONNX
  • Leveraging ONNX Runtime for GPU/WebGL/TensorRT/DirectML offloading
  • Combining multiple image processing operations into a single distributable ONNX model

Requirements

Python 3.10 or later

numpy          2.4.2     or later
onnx           1.20.1    or later
onnxruntime    1.24.2    or later
opencv-python  4.13.0.92 or later
pyvis          0.3.2     or later
pytest         9.0.2     or later
scipy          1.17.1    or later

Installation

# Clone repository
git clone https://github.com/Kazuhito00/onnx-cv-graph
cd onnx-cv-graph

# Install Python packages
pip install -r requirements.txt

ONNX Export

Export and Test ONNX Models

Run the following script to export all ONNX files.
Pre-tested ONNX files are already included in the repository, so this step is not required if you only want to run inference.

python src/export_all.py
python -m pytest tests/ -v

See TEST_DESIGN.md for testing guidelines.

Model List

A list of implemented and planned image processing operations is available in MODELS.md.

Usage

Python Example

Grayscale conversion

python example_grayscale.py

Grayscale conversion using ChainOp (single ONNX export)

python example_grayscale_chainop.py

Web App Demo using ONNX Runtime Web

python -m http.server 8080
# http://localhost:8080/example_app.html

Project Structure

README.md                # README (Japanese)
README_EN.md             # README (English)
MODELS.md                # Implemented / planned image processing list
TEST_DESIGN.md           # Testing guidelines
requirements.txt         # Python dependencies
example_app.html         # onnxruntime-web demo
src/
  base.py                # OnnxGraphOp abstract class
  chain.py               # ChainOp (serial composition of multiple ops)
  export_all.py          # Batch export + models_meta.json generation
  onnx_cv_graph/         # Graph implementations
models/                  # ONNX file storage
tests/                   # pytest test cases
assets/                  # Sample images and graph visualization HTML

Author

Kazuhito Takahashi (https://x.com/KzhtTkhs)

License

onnx-cv-graph is under Apache-2.0 license.

License (Image)

The sample image uses "猫背が治った！" from PAKUTASO.