README_en.md

简体中文 | English

PaddleVideo

Introduction

PaddleVideo is a toolset for video tasks prepared for the industry and academia. This repository provides examples and best practice guildelines for exploring deep learning algorithm in the scene of video area.

Update:

• release 🔥PP-TSMv2, an lite action recognition model, top1_acc on Kinetics-400 is 74.38%，cpu inference time on 10s video with 25fps is only 433ms. benchmark.
• add Knowledge Distilltion framework code.
• add TokenShift, 2s-ACGN and CTR-GCN model.

​ 💖 Welcome to scan the code and join the group discussion 💖

• Scan the QR code below with your Wechat and reply "video", you can access to official technical exchange group. Look forward to your participation.

Features

PaddleVideo support a variety of cutting-edge algorithms related to video, and developed industrial featured models/solution PP-TSM and PP-TSMv2 on this basis, and get through the whole process of data production, model training, compression, inference and deployment.

Quick Start

• One line of code quick use: Quick Start

Tutorials

• Quick Start
• Installation
• Usage
• PP-TSM🔥
  • Model Zoo
  • Model training
  • Model Compression
    • Model Quantization
    • Knowledge Distillation
  • Inference and Deployment
    • Python Inference
    • C++ Inference
    • Serving
    • Paddle2ONNX
    • Benchmark
• Academic algorithms🚀
• Datasets
• Data Annotation
• Contribute

License

PaddleVideo is released under the Apache 2.0 license.