πŸš€ PocketPose Model Zoo – Free Pre-Trained Models for Human Pose Estimation

Welcome to the PocketPose Model Zoo, your ultimate resource for pre-trained models designed for fast, accurate, and efficient human pose estimation across a variety of frameworks and keypoint formats. Whether you’re building real-time applications or conducting cutting-edge research, our models provide high-performance solutions without the hassle of training from scratch.

For minimal setup with automatic selection of the best model for your use case, visit our Get Started page.

For Researchers & Developers: Alongside the latest high-performance pose estimation models, we provide access to legacy models that played a pivotal role in shaping the field. These older models are ideal for comparative research, benchmarking, and academic exploration.

Use our interactive Model Explorer below to filter and discover all available models in the PocketPose Model Zoo.

* Reported Metrics: COCO AP for 2D models using COCO/COCO-WholeBody keypoints, PCKh for 2D models with MPII keypoints, and MPJPE for 3D models. Metrics are evaluated on validation datasets using the same person detector (RTMDet) where applicable. Higher values indicate better performance.
** Speed: Measured on a 2024 MacBook Pro 14” (M4 Pro) using ONNX Runtime. Lower values indicate better performance.
Name Input Size Keypoints Size (MB) Accuracy* Speed (FPS)** Backend Source License
RTMW (2024)
Top-down | CSPNeXt | Coordinate Classification
RTMW-m 256x192x3 COCO_WHOLEBODY ONNX MMPose: Code · Weights Apache-2.0
RTMPose (2023)
Top-down | CSPNeXt | Coordinate Classification
RTMPose-t 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
RTMPose-s 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
RTMPose-m 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
RTMPose-l 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
SimCC (2021)
Top-down | ResNet, MobileNetV2, ViPNAS-MBV3 | Coordinate Classification
SimCC-mbv2 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
SimCC-ViPNAS-MBV3 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
SimCC-ResNet50 256x192x3 COCO ONNX MMDeploy: Code · Weights Apache-2.0
MoveNet (2021)
Bottom-up | MobileNetV2 + FPN | Person Center, Keypoint Heatmap, Keypoint Regression, and Offset Heads
MoveNet Lightning 192x192x3 COCO TFLITE TensorFlow Hub: Weights Apache-2.0
MoveNet Lightning FP16 192x192x3 COCO TFLITE TensorFlow Hub: Weights Apache-2.0
MoveNet Lightning INT8 192x192x3 COCO TFLITE TensorFlow Hub: Weights Apache-2.0
MoveNet Thunder 256x256x3 COCO TFLITE TensorFlow Hub: Weights Apache-2.0
MoveNet Thunder FP16 256x256x3 COCO TFLITE TensorFlow Hub: Weights Apache-2.0
MoveNet Thunder Int8 256x256x3 COCO TFLITE TensorFlow Hub: Weights Apache-2.0
EfficientPose (2020)
Bottom-up | EfficientNet | Heatmap and Part Affinity Fields Regression
EfficientPose-RT 224x224x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-RT-Lite 224x224x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-I 256x256x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-I-Lite 256x256x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-II 368x368x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-II-Lite 368x368x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-III 480x480x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose-IV 600x600x3 MPII TFLITE Official Repo: Code · Weights Apache-2.0
EfficientPose with NAS (2020)
Top-down | NAS-Optimized Backbone | Heatmap Regression
NASNet-A (COCO) 256x192x3 COCO TFLITE Official Repo: Code · Weights Unspecified
NASNet-B (COCO) 256x192x3 COCO TFLITE Official Repo: Code · Weights Unspecified
NASNet-B (MPII) 256x256x3 MPII TFLITE Official Repo: Code · Weights Unspecified
NASNet-C (COCO) 256x192x3 COCO TFLITE Official Repo: Code · Weights Unspecified
NASNet-C (MPII) 256x256x3 MPII TFLITE Official Repo: Code · Weights Unspecified
BlazePose (2020)
Top-down | 1.5-Stage Tiny Hourglass | Direct Keypoint Regression
BlazePose-Lite 256x256x3 BLAZEPOSE MEDIAPIPE MediaPipe: Code · Weights Apache-2.0
BlazePose-Full 256x256x3 BLAZEPOSE MEDIAPIPE MediaPipe: Code · Weights Apache-2.0
BlazePose-Heavy 256x256x3 BLAZEPOSE MEDIAPIPE MediaPipe: Code · Weights Apache-2.0
Lightweight Pose Network (2019)
Top-down | ResNet + Lightweight Bottleneck Blocks | Heatmap Regression
LPN-50 (COCO) 256x192x3 COCO ONNX Official Repo: Code · Weights MIT
LPN-101 (COCO) 256x192x3 COCO ONNX Official Repo: Code · Weights MIT
LPN-152 (COCO) 256x192x3 COCO ONNX Official Repo: Code · Weights MIT
LPN-50 (MPII) 256x256x3 MPII ONNX Official Repo: Code · Weights MIT