object detection / instance segmentation / multi-object tracking / pose estimation
PaddleDetection is a PaddlePaddle-based toolkit for object detection, instance segmentation, multi-object tracking, and keypoint detection.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
PyTorch DETR-centric research platform — choose if you need DETR-family models in PyTorch, not the breadth of PaddleDetection.
detrex is a research platform for DETR-based object detection, segmentation, pose estimation and other visual recognition tasks.
Mature PyTorch toolbox with a large model zoo — choose for PyTorch-centric workflows, though less actively maintained now.
OpenMMLab Detection Toolbox and Benchmark
MXNet Gluon CV toolkit with detection/segmentation — choose if you need MXNet, but it is not actively developed.
Gluon CV Toolkit
General mobile-friendly CV training library that includes detection — choose for a broader CV training abstraction.
CVNets: A library for training computer vision networks
State-of-the-art PyTorch toolkit covering detection, segmentation, pose, and tracking — choose for Ultralytics' API, but mind AGPL license.
Ultralytics YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Specific real-time DETR-family model from Roboflow — choose for fine-tuning that architecture, not a full toolkit.
RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. [ICLR 2026]
Ultra-lightweight anchor-free detector — choose for mobile/edge deployment where PaddleDetection models are too heavy.
NanoDet-Plus⚡Super fast and lightweight anchor-free object detection model. 🔥Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphone🔥
Research implementation of a single segmentation model — choose to reproduce Mask2Former, not for a toolkit.
Code release for "Masked-attention Mask Transformer for Universal Image Segmentation"
Research code for SOLO instance segmentation — choose for that exact architecture.
SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.
Research implementation of the Mask DINO transformer — narrow, not a general toolkit.
[CVPR 2023] Official implementation of the paper "Mask DINO: Towards A Unified Transformer-based Framework for Object Detection and Segmentation"
Old anchor-free instance segmentation model implementation — choose only for that model.
[CVPR 2020] CenterMask : Real-Time Anchor-Free Instance Segmentation
Another CenterMask version, still a single-model implementation — limited to that architecture.
[CVPR 2020] CenterMask : Real-time Anchor-Free Instance Segmentation
Minimal PyTorch detection framework with classic models — choose for quick experiments without CUDA build.
一个简单方便的目标检测框架(PyTorch环境可直接运行，不需要cuda编译)，支持Faster_RCNN、Cascade_RCNN、Yolo系列、SSD等经典网络。
Multi-camera tracking using YOLO detectors — choose if you need just tracking with StrongSORT.
Real-time multi-camera multi-object tracker using YOLO varients
Unofficial YOLOv7-style detector with transformers — choose for that specific architecture or TensorRT.
🔥🔥🔥🔥 (Earlier YOLOv7 not official one) YOLO with Transformers and Instance Segmentation, with TensorRT acceleration! 🔥🔥🔥
Older PyTorch Mask R-CNN reimplementation — use if you specifically need Mask R-CNN in PyTorch.
Mask R-CNN for object detection and instance segmentation on Pytorch
Multi-version YOLO+DeepSORT tracking pipeline — choose for PyTorch detection+tracking.
Deepsort with yolo series. This project support the existing yolo detection model algorithm (YOLOV8, YOLOV7, YOLOV6, YOLOV5, YOLOV4Scaled, YOLOV4, YOLOv3', PPYOLOE, YOLOR, YOLOX ).
YOLOX+DeepSORT tracker — narrow, for that specific combination.
using yolox+deepsort for object-tracking
Keras/TensorFlow Mask R-CNN — choose for TensorFlow deployment of that single model.
Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
YOLOv5+DeepSORT example — archived, only a reference implementation.
DeepSORT + YOLOv5
Skeleton-based action recognition and tracking — choose for that sub-task, not general detection.
Multi Person Skeleton Based Action Recognition and Tracking
Ready-to-use YOLOv5+DeepSORT inference with counting — choose for a quick tracking/counting script.
Yolov5 deepsort inference，使用YOLOv5+Deepsort实现车辆行人追踪和计数，代码封装成一个Detector类，更容易嵌入到自己的项目中
YOLOX+DeepSORT baseline tracker — narrow, for that baseline.
基于YoloX目标检测+DeepSort算法实现多目标追踪Baseline
Tracking-only library that can wrap any detector — choose when you already have a detector and need a lightweight tracker.
Lightweight Python library for adding real-time multi-object tracking to any detector.
These projects were analysed and named PaddleDetection among their alternatives. The relationship is not symmetric — how PaddleDetection rates them is a separate judgement, made when PaddleDetection is analysed in its own right.
calls PaddleDetection “choose if you are in the PaddlePaddle ecosystem and need Apache-2.0 with detection, tracking, and pose.”
Ultralytics YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking
calls PaddleDetection “Choose it when your stack is PaddlePaddle or you need Apache-2.0 plus multi-task detection/tracking/keypoint support; not a PyTorch drop-in.”
Ultralytics YOLOv5 in PyTorch for object detection, instance segmentation, classification, training, and export.
calls PaddleDetection “PaddlePaddle detection toolkit, choose if using Paddle ecosystem.”
OpenMMLab Detection Toolbox and Benchmark
calls PaddleDetection “Full-featured PaddlePaddle toolkit for detection/segmentation; choose for active development, deployment, and PaddlePaddle compatibility.”
Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
calls PaddleDetection “Choose when you prefer a maintained PaddlePaddle framework and need many detector/segmentation/tracking models rather than one YOLOv7 PyTorch repo.”
Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
calls PaddleDetection “Use it if you prefer PaddlePaddle and need a comprehensive detection toolbox with many architectures.”
YOLOv10: Real-Time End-to-End Object Detection [NeurIPS 2024]
calls PaddleDetection “Choose when you want a broader object-detection toolbox under Apache-2.0 and the PaddlePaddle framework.”
PyTorch implementation of YOLOv3, YOLOv3-SPP, and YOLOv3-tiny for real-time object detection with training, validation, inference, and multi-format export.
calls PaddleDetection “choose it when you are already in the PaddlePaddle ecosystem and need detection, segmentation, tracking, and pose models from one toolkit.”
RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. [ICLR 2026]
calls PaddleDetection “Choose when you want an Apache-2.0 PaddlePaddle toolkit with many detection/segmentation/pose models instead of a single standalone MIT PyTorch CenterNet repo.”
Object detection, 3D detection, and pose estimation using center point detection:
calls PaddleDetection “Pick it if you are already on PaddlePaddle and want a full detection/segmentation/tracking toolkit, accepting fixed classes instead of text prompts.”
[CVPR 2024] Real-Time Open-Vocabulary Object Detection
calls PaddleDetection “Choose when you're tied to the PaddlePaddle ecosystem or need its deployment tooling for instance segmentation.”
SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.
calls PaddleDetection “Choose if you are standardized on PaddlePaddle and need a current detection/segmentation toolkit with deployment tools.”
Mask R-CNN for object detection and instance segmentation on Pytorch