lmk123568/PyVideoProc — Gitpedia

<p style="" align="center"> <img src="./assets/logo.png" alt="Logo" width="20%"> </p> <h1 align="center">PyVideoProc</h1> <p style="margin:0px" align="center"> <img src="https://img.shields.io/badge/license-BSD--2-blue.svg?&logo=c&logoColor=white&style=for-the-badge"> <img src="https://img.shields.io/badge/CUDA-12.8-76B900?&logo=nvidia&logoColor=white&style=for-the-badge"> <img src="https://img.shields.io/badge/OS-Linux-FCC624?&logo=linux&logoColor=white&style=for-the-badge"> <img src="https://img.shields.io/badge/ENV-Docker-2496ED?logo=docker&logoColor=white&style=for-the-badge"> <a href="https://youtu.be/-soVRjH1rb4" target="_blank"> <img src="https://img.shields.io/badge/Video-YouTube-FF0000?logo=youtube&logoColor=white&style=for-the-badge"> </a> <a href="https://www.bilibili.com/video/BV12TcvzbEcZ" target="_blank"> <img src="https://img.shields.io/badge/Video-Bilibili-FB7299?logo=bilibili&logoColor=white&style=for-the-badge"> </a> </p>

PyVideoProc 提供基于 CUDA 加速优化的高性能 Python SDK，可高效实现多路、多卡、多模型的视频解码、AI 推理与编码，显著降低开发复杂度并提升吞吐性能

PyVideoProc provides a high-performance Python SDK optimized with CUDA acceleration, enabling efficient multi-stream, multi-GPU, and multi-model video decoding, AI inference, and encoding—significantly reducing development complexity and boosting throughput performance.

https://github.com/user-attachments/assets/402f9080-004e-457e-8d36-3fefdb56f21d

⭐ 多进程绕过 GIL 限制，提升 Python 并发性能

⭐ 减少 Host-Device 数据传输，降低 GPU 显存冗余拷贝，提升推理速度

⭐ 尽可能在 GPU 上计算，以降低 CPU 计算负担

⭐ 开箱即用，简单易懂，扩展性强，适合中小型项目快速部署

	Open Source	Learning Curve	Developer Friendliness	Performance	Architecture Design
DeepStream	❌	High	Low	High	Single-process, multi-threaded
VideoPipe	✅	medium（requires cpp knowledge）	Medium（requires cpp knowledge）	Medium	Single-process, multi-threaded
Our	✅	≈ 0	High	Medium	Multi-process, single-threaded

🔗 Bilibili: https://www.bilibili.com/video/BV12TcvzbEcZ

🔗 YouTube: https://youtu.be/-soVRjH1rb4

Quick Start

本项目推荐 Docker 容器运行，首先确保本地环境满足以下三个条件：

Docker >= 24.0.0
NVIDIA Driver >= 590
NVIDIA Container Toolkit >= 1.13.0

1. 生成镜像

clone 本项目，生成包含完整开发环境的镜像

bash
git clone https://github.com/lmk123568/PyVideoProc.git
cd PyVideoProc/docker
docker build -t pyvideoproc:cuda12.8 .

镜像生成后，进入容器，不报错即成功

bash
docker run -it \
  --privileged \
  --gpus all \
  -e NVIDIA_DRIVER_CAPABILITIES=all \
  -v {your_path}/PyVideoProc:/workspace \
  pyvideoproc:cuda12.8 \
  bash

后续示例代码默认在容器内/workspace运行

⚠️ 不推荐自己本地装环境，如果一定要自己装，请参考 Dockerfile

2. 编译加速包

bash
python install .
# python install -e .  # 开发模式

这里面包含了硬件编解码、YOLO26 推理优化的 C++ 实现，并通过 Pybind11 给 Python 调用

3. 训练模型权重转换

将通过 ultralytics 训练的pt模型导入到当前目录（/workspace）下（示例模型为 yolo26n.pt）

bash
python pt2trt.py  --w ./yolo26n.pt --fp16

转换过程中会与 ultralytics 官方结果进行推理对齐

💡 TensorRT 编译生成 .engine 过程中，推理尺寸默认设置为(576,1024)，可以跳过letterbox降低计算开销

💡 遇到警告 requirements: Ultralytics requirement ['onnxruntime-gpu'] not found ... 可以 Ctrl + C 跳过

4. 运行

开启 MPS（Multi-Process Service）

bash
nvidia-cuda-mps-control -d
# echo quit | nvidia-cuda-mps-control  关闭 MPS

阅读理解其代码并运行

bash
python main.py

Benchmark

测试日期: 2026-01-25

测试硬件: AMD Ryzen 9 5950 X + NVIDIA GeForce RTX 3090

测试任务: 4 × RTSP Decoders → YOLO26 (TensorRT) → 4 × RTMP Encoders

	CPU	RAM	GPU VRAM	GPU-Util
VideoPipe（ffmpeg codec）	511.6 %	1.5 GiB	2677 MiB	16 %
Our	40 %	1.2GiB	3932 MiB	12 %

License

BSD-2-Clause