object detection / 3D detection / pose estimation (center-point keypoint method)
CenterNet performs object detection, monocular 3D detection, and pose estimation by treating each object as a single center point and regressing additional attributes from that keypoint.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
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 toolkit based on PaddlePaddle. It supports object detection, instance segmentation, multiple object tracking and real-time multi-person keypoint detection.
Choose when you need open-vocabulary detection driven by text prompts; CenterNet only predicts fixed training classes.
[CVPR 2024] Real-Time Open-Vocabulary Object Detection
Choose when you want a continuously maintained YOLO framework for detection/segmentation/pose with easy pip workflow, but note AGPL-3.0 license vs MIT and a different model family.
Ultralytics YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking
Choose when your need is specifically real-time multi-person 2D body/face/hand pose; bottom-up PAF output is more pose-focused and does not give object boxes/3D.
OpenPose: Real-time multi-person keypoint detection library for body, face, hands, and foot estimation
Choose when you want a modular benchmark toolbox with many detector implementations and reproducible configs, using the MMDetection ecosystem rather than CenterNet's single model.
OpenMMLab Detection Toolbox and Benchmark
Choose when you want transformer-based end-to-end detection with no anchors or NMS; code is Apache-2.0 but unmaintained/archived.
End-to-End Object Detection with Transformers
Choose when you want a real-time GPL-3.0 YOLOv7 model with proven deployment, not the center-point/pose/3D model.
Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
Choose when you need an actively developed real-time DETR architecture for fine-tuning detection/segmentation; Apache-2.0 is permissive.
RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. [ICLR 2026]
Choose when you want Meituan's GPL-3.0 YOLOv6 for industrial edge deployment and need no pose/3D outputs.
YOLOv6: a single-stage object detection framework dedicated to industrial applications.
Choose when you want PaddlePaddle implementations of YOLO/RT-DETR and prefer staying in that ecosystem instead of the standalone CenterNet repo.
🚀🚀🚀 YOLO series of PaddlePaddle implementation, PP-YOLOE+, RT-DETR, YOLOv5, YOLOv6, YOLOv7, YOLOv8, YOLOv10, YOLO11, YOLOX, YOLOv5u, YOLOv7u, YOLOv6Lite, RTMDet and so on. 🚀🚀🚀
Choose when you want a complete multi-person 2D pose-and-tracking system with an integrated detector instead of CenterNet's center-based component.
Real-Time and Accurate Full-Body Multi-Person Pose Estimation&Tracking System
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Choose when you need a CPU-lightweight 2D pose model and can accept an old unmaintained codebase that lacks 3D/object detection.
Fast and accurate human pose estimation in PyTorch. Contains implementation of "Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose" paper.
Choose when you need high-resolution bottom-up 2D pose only; unmaintained (>5y) and no detection/3D.
This is an official implementation of our CVPR 2020 paper "HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation" (https://arxiv.org/abs/1908.10357)
Choose when you need TensorRT-accelerated pose running on Jetson/edge rather than a trainable PyTorch CenterNet; old code and no object detection.
Real-time pose estimation accelerated with NVIDIA TensorRT
Choose when you specifically want the other 'CenterNet: Keypoint Triplets' detector for 2D object detection, accepting a stale repo with none of the 3D/pose extensions.
Codes for our paper "CenterNet: Keypoint Triplets for Object Detection" .
Choose when you need an older PyTorch toolbox of 2D pose architectures; it is narrower, unmaintained, and won't detect objects.
A PyTorch toolkit for 2D Human Pose Estimation.
Choose when you need the top-down Simple Baselines 2D pose stage; it is archived and assumes person boxes rather than end-to-end detection.
The project is an official implement of our ECCV2018 paper "Simple Baselines for Human Pose Estimation and Tracking(https://arxiv.org/abs/1804.06208)"
Choose when you need the legacy TensorFlow Cascaded Pyramid Network for 2D pose; unmaintained, no object detection/3D.
[CVPR 2018] Cascaded Pyramid Network for Multi-Person Pose Estimation
These projects were analysed and named CenterNet among their alternatives. The relationship is not symmetric — how CenterNet rates them is a separate judgement, made when CenterNet is analysed in its own right.
calls CenterNet “Choose for an MIT-licensed center-point/heatmap detector if you want no anchor priors, accepting it is much older/unmaintained.”
Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
calls CenterNet “Choose it if you need a keypoint-based center detection alternative with simple post-processing.”
YOLOv10: Real-Time End-to-End Object Detection [NeurIPS 2024]
calls CenterNet “Choose if you prefer a keypoint-based detector with MIT license and a pure-PyTorch codebase.”
PyTorch implementation of YOLOv3, YOLOv3-SPP, and YOLOv3-tiny for real-time object detection with training, validation, inference, and multi-format export.
calls CenterNet “Center-based detection and pose estimation; a different approach useful for multi-task models on embedded devices.”
Real-Time and Accurate Full-Body Multi-Person Pose Estimation&Tracking System
calls CenterNet “choose for center-point detection research code, but it is unmaintained.”
Ultralytics YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking
calls CenterNet “Center-based object detector with pose estimation; choose for detection/pose tasks, but no instance segmentation.”
Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
calls CenterNet “choose it if you need a simple keypoint-based detection/pose/3D baseline and accept an old, unmaintained codebase.”
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 CenterNet “Only for legacy center/keypoint-based detection code; unmaintained for over three years and class-limited.”
[CVPR 2024] Real-Time Open-Vocabulary Object Detection