Skip to content

fjyu95/vision-engine

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

vision-engine

A high-performance vision inference engine with pipeline orchestration, multi-GPU scheduling, and modular algorithm integration.

项目目标

vision-engine 面向可扩展视觉算法工程化,采用模块化组织方式,支持按任务快速接入、验证和批量处理,当前包含:

  • 目标检测(Object Detection)
  • 图像去噪(Denoise)
  • 超分辨率(Super Resolution)
  • 切片压缩(Tiling Compress)
  • 模板模块(Template,用于新算法快速接入)

目录结构

vision-engine/
├── models/
│   ├── denoise/
│   ├── object-detection/
│   ├── super-resolution/
│   ├── tiling-compress/
│   └── template/
├── pyproject.toml
├── setup.py
├── requirements.txt
└── README.md

各算法目录通常包含:

  • README.md:模块说明与使用方式
  • requirements.txt:模块依赖
  • *-demo.py / test_batch_*.py:单图与批处理示例
  • algorithms.md:模型/算法选型与说明(按模块提供)

环境准备

当前推荐安装流程:

conda create -n vision-engine python=3.10
conda activate vision-engine
pip install torch==2.7.1 torchvision --index-url https://download.pytorch.org/whl/cu128
pip install -r requirements.txt

兼容性备注:

torch260+cu124+flash273
torch280+cu128+flash283

使用方式(初始版)

建议按模块进入对应目录运行示例脚本,例如:

cd models/object-detection
python yolo_detection-demo.py

或执行批处理脚本(以模块内 README 为准)。

开发约定(初始)

  • 新增算法优先放入 models/<task-name>/
  • 目录内保持统一结构:README + requirements + demo + batch script
  • 模块依赖尽量局部化,避免影响全局环境
  • 先保证可运行,再逐步补充性能优化与工程化接口

About

A high-performance vision inference engine with pipeline orchestration, multi-GPU scheduling, and modular algorithm integration.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages