README_en.md

English | 简体中文

PaddleDetection 2.0 is ready! Dygraph mode is set by default and static graph code base is here

Keypoint detection and Multi-Object Tracking are released!

Highly effective PPYOLO v2 and ultra lightweight PPYOLO tiny are released! link

SOTA Anchor Free model -- PAFNet is released! link

Introduction

PaddleDetection is an end-to-end object detection development kit based on PaddlePaddle, which aims to help developers in the whole development of constructing, training, optimizing and deploying detection models in a faster and better way.

PaddleDetection implements varied mainstream object detection algorithms in modular design, and provides wealthy data augmentation methods, network components(such as backbones), loss functions, etc., and integrates abilities of model compression and cross-platform high-performance deployment.

After a long time of industry practice polishing, PaddleDetection has had smooth and excellent user experience, it has been widely used by developers in more than ten industries such as industrial quality inspection, remote sensing image object detection, automatic inspection, new retail, Internet, and scientific research.

Product news

• 2021.05.20: Release release/2.1 version. Release Keypoint Detection, including HigherHRNet and HRNet, Multi-Object Tracking, including DeepSORT，JDE and FairMOT. Release model compression for PPYOLO series models.Update documents such as EXPORT ONNX MODEL. Please refer to PaddleDetection for details.
• 2021.04.14: Release release/2.0 version. Dygraph mode in PaddleDetection is fully supported. Cover all the algorithm of static graph and update the performance of mainstream detection models. Release PP-YOLO v2 and PP-YOLO tiny, enhanced anchor free model PAFNet and S2ANet which is aimed at rotation object detection.Please refer to PaddleDetection for details.
• 2020.02.07: Release release/2.0-rc version, Please refer to PaddleDetection for details.

Features

• Rich Models PaddleDetection provides rich of models, including 100+ pre-trained models such as object detection, instance segmentation, face detection etc. It covers a variety of global competition champion schemes.

• Highly Flexible: Components are designed to be modular. Model architectures, as well as data preprocess pipelines and optimization strategies, can be easily customized with simple configuration changes.

• Production Ready: From data augmentation, constructing models, training, compression, depolyment, get through end to end, and complete support for multi-architecture, multi-device deployment for cloud and edge device.

• High Performance: Based on the high performance core of PaddlePaddle, advantages of training speed and memory occupation are obvious. FP16 training and multi-machine training are supported as well.

Overview of Kit Structures

Architectures	Backbones	Components	Data Augmentation
Two-Stage Detection Faster RCNN FPN Cascade-RCNN Libra RCNN Hybrid Task RCNN PSS-Det RCNN One-Stage Detection RetinaNet YOLOv3 YOLOv4 PP-YOLO SSD Anchor Free CornerNet-Squeeze FCOS TTFNet Instance Segmentation Mask RCNN SOLOv2 Face-Detection FaceBoxes BlazeFace BlazeFace-NAS	ResNet(&vd) ResNeXt(&vd) SENet Res2Net HRNet Hourglass CBNet GCNet DarkNet CSPDarkNet VGG MobileNetv1/v3 GhostNet Efficientnet	Common Sync-BN Group Norm DCNv2 Non-local FPN BiFPN BFP HRFPN ACFPN Loss Smooth-L1 GIoU/DIoU/CIoU IoUAware Post-processing SoftNMS MatrixNMS Speed FP16 training Multi-machine training	Resize Flipping Expand Crop Color Distort Random Erasing Mixup Cutmix Grid Mask Auto Augment

Overview of Model Performance

The relationship between COCO mAP and FPS on Tesla V100 of representative models of each architectures and backbones.

NOTE:

• CBResNet stands for Cascade-Faster-RCNN-CBResNet200vd-FPN, which has highest mAP on COCO as 53.3%

• Cascade-Faster-RCNN stands for Cascade-Faster-RCNN-ResNet50vd-DCN, which has been optimized to 20 FPS inference speed when COCO mAP as 47.8% in PaddleDetection models

• PP-YOLO achieves mAP of 45.9% on COCO and 72.9FPS on Tesla V100. Both precision and speed surpass YOLOv4

• PP-YOLO v2 is optimized version of PP-YOLO which has mAP of 49.5% and 68.9FPS on Tesla V100

• All these models can be get in Model Zoo

Tutorials

Get Started

• Installation guide
• Prepare dataset
• Quick start on PaddleDetection

Advanced Tutorials

• Parameter configuration

  • Parameter configuration for RCNN model
  • Parameter configuration for PP-YOLO model

• Model Compression(Based on PaddleSlim)

  • Prune/Quant/Distill

• Inference and deployment

  • Export model for inference
  • Paddle Inference
    • Python inference
    • C++ inference
  • Paddle-Lite
  • Paddle Serving
  • Export ONNX model
  • Inference benchmark

• Advanced development

  • New data augmentations
  • New detection algorithms

Model Zoo

• Universal object detection
  • Model library and baselines
  • PP-YOLO
  • Enhanced Anchor Free model--TTFNet
  • Mobile models
  • 676 classes of object detection
  • Two-stage practical PSS-Det
  • SSLD pretrained models
• Universal instance segmentation
  • SOLOv2
• Rotation object detection
  • S2ANet
• Keypoint detection
  • HigherHRNet
  • HRNet
• Multi-Object Tracking
  • DeepSORT
  • JDE
  • FairMOT
• Vertical field
  • Face detection
  • Pedestrian detection
  • Vehicle detection
• Competition Plan
  • Objects365 2019 Challenge champion model
  • Best single model of Open Images 2019-Object Detection

Applications

• Christmas portrait automatic generation tool

Updates

v2.2 was released at 08/2021, release Transformer detection models, release Dark HRNet keypoint detection model, release tracking models of head and vehicle, release optimized S2ANet model, inference with batch size > 1 supported for main architectures. Please refer to change log for details.

v2.1 was released at 05/2021, Release Keypoint Detection and Multi-Object Tracking. Release model compression for PPYOLO series. Update documents such as export ONNX model. Please refer to change log for details.

v2.0 was released at 04/2021, fully support dygraph version, which add BlazeFace, PSS-Det and plenty backbones, release PP-YOLOv2, PP-YOLO tiny and S2ANet, support model distillation and VisualDL, add inference benchmark, etc. Please refer to change log for details.

License

PaddleDetection is released under the Apache 2.0 license.

Contributing

Contributions are highly welcomed and we would really appreciate your feedback!!

• Thanks Mandroide for cleaning the code and unifying some function interface.
• Thanks FL77N for contributing the code of Sparse-RCNN model.

Citation

@misc{ppdet2019,
title={PaddleDetection, Object detection and instance segmentation toolkit based on PaddlePaddle.},
author={PaddlePaddle Authors},
howpublished = {\url{https://github.com/PaddlePaddle/PaddleDetection}},
year={2019}
}