sapiens2-pose-5b

Sapiens2-5B-Pose

308-keypoint top-down pose estimation including detailed face (274 keypoints), hand, and foot keypoints. Predictions follow the Sociopticon keypoint format.

This repository contains the 5B Pose Estimation checkpoint, finetuned from the Sapiens2-5B pretrained backbone.

Pose is top-down — it requires bounding boxes from a person detector. We use RTMDet.

• 📄 Paper: arXiv:2604.21681
• 🌐 Project Page: rawalkhirodkar.github.io/sapiens2
• 💻 Code: github.com/facebookresearch/sapiens2

Model Details

• Developed by: Meta
• Model type: Vision Transformer
• License: Sapiens2 License
• Task: pose
• Base model: facebook/sapiens2-pretrain-5b
• Format: safetensors
• File: sapiens2_5b_pose.safetensors

Quick Start

Install the Sapiens2 repo (pip install -e .), download the checkpoint, and run the demo:

# 1. Download the checkpoint to $SAPIENS_CHECKPOINT_ROOT/pose/
hf download facebook/sapiens2-pose-5b sapiens2_5b_pose.safetensors \
    --local-dir ~/sapiens2_host/pose

# 2. Run the demo (edit INPUT, OUTPUT, and MODEL_NAME inside the script)
cd $SAPIENS_ROOT/sapiens/pose
./scripts/demo/keypoints308.sh

See the Pose Estimation guide for details on inputs, outputs, and visualization options.

Model Card

Sapiens2-Pose Family

See the Sapiens2 Collection for all variants and other downstream task checkpoints.

Intended Use

• Pose Estimation on human-centric imagery
• Research on human-centric vision

License

Released under the Sapiens2 License.

Citation

@article{khirodkarsapiens2,
  title={Sapiens2},
  author={Khirodkar, Rawal and Wen, He and Martinez, Julieta and Dong, Yuan and Su, Zhaoen and Saito, Shunsuke},
  journal={arXiv preprint arXiv:2604.21681},
  year={2026}
}