14.4K ·GitHub
What is this?
PaddleDetection is an object detection toolkit built on the PaddlePaddle deep learning framework, covering object detection, instance segmentation, multi-object tracking, and keypoint detection.
PaddleDetection is an object detection toolkit built on the PaddlePaddle deep learning framework, covering object detection, instance segmentation, multi-object tracking, and keypoint detection.Is it a framework or a collection of pretrained models?It's both: a modular framework with decoupled components (backbones, necks, losses, data augmentation) plus a large library of pretrained models and industry-specific solutions like PP-YOLOE, PP-PicoDet, and PP-TinyPose.
It's both: a modular framework with decoupled components (backbones, necks, losses, data augmentation) plus a large library of pretrained models and industry-specific solutions like PP-YOLOE, PP-PicoDet, and PP-TinyPose.Do I need the PaddlePaddle framework first?Yes, it is built entirely on PaddlePaddle, Baidu's deep learning framework, which must be installed before using this toolkit.
Yes, it is built entirely on PaddlePaddle, Baidu's deep learning framework, which must be installed before using this toolkit.Is the README fully available in English?The README is bilingual (Simplified Chinese | English) but most of the detailed sections provided are in Chinese, so English speakers may need to rely on translated docs or the English README toggle.
The README is bilingual (Simplified Chinese | English) but most of the detailed sections provided are in Chinese, so English speakers may need to rely on translated docs or the English README toggle.Can models be deployed outside of servers, like on edge devices?Yes, it supports server-side, mobile/edge deployment via Paddle Lite, Paddle Serving, C++ and Python inference, and ONNX export.
Yes, it supports server-side, mobile/edge deployment via Paddle Lite, Paddle Serving, C++ and Python inference, and ONNX export.What are PP-YOLOE and PP-PicoDet?PP-YOLOE is a high-accuracy real-time object detection model, and PP-PicoDet is an ultra-lightweight real-time detector optimized for mobile and edge deployment.
PP-YOLOE is a high-accuracy real-time object detection model, and PP-PicoDet is an ultra-lightweight real-time detector optimized for mobile and edge deployment.What is PP-Human and PP-Vehicle?PP-Human is a real-time pedestrian analysis toolkit (behavior, attributes, crowd counting) and PP-Vehicle is a real-time vehicle analysis toolkit (license plate recognition, traffic flow, violation detection).
PP-Human is a real-time pedestrian analysis toolkit (behavior, attributes, crowd counting) and PP-Vehicle is a real-time vehicle analysis toolkit (license plate recognition, traffic flow, violation detection).Which tasks does the model zoo cover?It covers 2D/3D object detection, instance segmentation, face detection, keypoint/pose detection, multi-object tracking, and semi-supervised learning, with models like Faster RCNN, YOLOv3/v5/v7/X, RT-DETR, Mask-RT-DETR, FairMOT, and DeepSORT.
It covers 2D/3D object detection, instance segmentation, face detection, keypoint/pose detection, multi-object tracking, and semi-supervised learning, with models like Faster RCNN, YOLOv3/v5/v7/X, RT-DETR, Mask-RT-DETR, FairMOT, and DeepSORT.Why would I use this instead of plain YOLO repos?PaddleDetection bundles dozens of SOTA detectors, trackers, and keypoint models under one modular framework with industry-ready pipelines and multi-hardware deployment support (NVIDIA GPU, Kunlun, Ascend, Cambricon, Hygon), rather than a single model implementation.
PaddleDetection bundles dozens of SOTA detectors, trackers, and keypoint models under one modular framework with industry-ready pipelines and multi-hardware deployment support (NVIDIA GPU, Kunlun, Ascend, Cambricon, Hygon), rather than a single model implementation.What are similar projects?mmdetection, yolov7, YOLOX, yolov5, X-AnyLabeling, deep-learning-for-image-processing, Swin-Transformer, yolov3
mmdetection, yolov7, YOLOX, yolov5, X-AnyLabeling, deep-learning-for-image-processing, Swin-Transformer, yolov3Which topics is it in?mask-rcnn, instance-segmentation, face-detection, object-detection, onnx, computer-vision, china, pytorch, cloud, mobile, deep-learning, machine-learning
mask-rcnn, instance-segmentation, face-detection, object-detection, onnx, computer-vision, china, pytorch, cloud, mobile, deep-learning, machine-learning