DEKR

DEKR estimates multi-person poses from image-wide keypoint heatmaps and offsets.

Tasks: pose
Sizes: w32 at 640 px
Install: pip install libreyolo
Support tier: Inference only, since v1.6.0. Predict, validate and export only. Training features do not apply.
Upstream: DEKR by HRNet authors, MIT. Paper, source
Licenses: Code MIT, weights No separate artifact grant recorded. Commercial use

Install

bash

pip install "libreyolo"

Predict

Python

from libreyolo import LibreYOLO, SAMPLE_IMAGE # Downloads the W32 checkpoint from the upstream CDN and checks its SHA-256model = LibreYOLO("LibreDEKRw32-pose.pt", device="cpu")result = model(SAMPLE_IMAGE)print(result.keypoints)

The adapter uses the W32 no-deformable-convolution graph and 17 COCO person keypoints. Boxes enclose confident decoded joints. This is an inference-only family; the original deformable graph is a different checkpoint architecture.

Validate

Python

from libreyolo import LibreYOLO # Downloads the W32 checkpoint from the upstream CDN and checks its SHA-256model = LibreYOLO("LibreDEKRw32-pose.pt", device="cpu")# coco8-pose.yaml builds a 4-image COCO keypoint split on first usemetrics = model.val(data="coco8-pose.yaml", allow_download_scripts=True, workers=0)print(metrics)

Use the dataset format for this task. Validation explains the dataset requirements and returned metrics.

Export

Task	ONNX	TorchScript	ExecuTorch	TensorRT	OpenVINO	Paddle	MNN	RKNN	ncnn	TFLite	CoreML	Core AI
Pose	Pose to ONNX: supported	Pose to TorchScript: supported	Pose to ExecuTorch: not supported	Pose to TensorRT: supported	Pose to OpenVINO: supported	Pose to Paddle: not supported	Pose to MNN: not supported	Pose to RKNN: not supported	Pose to ncnn: not supported	Pose to TFLite: not supported	Pose to CoreML: not supported	Pose to Core AI: not supported

Python

from libreyolo import LibreYOLO # Downloads the W32 checkpoint from the upstream CDN and checks its SHA-256model = LibreYOLO("LibreDEKRw32-pose.pt", device="cpu")model.export(format="onnx")

Export setup lists format dependencies and loading exported artifacts.

Licensing

Check the license on the Hugging Face repository of the specific weights you download. Every checkpoint in the LibreYOLO org carries one, and they are not always the same across a family. That repository is the authoritative source; the summary below describes what applied when this page was last verified.

This is a description of the licenses involved, not legal advice. If the answer matters commercially, read the licenses yourself and take your own counsel.

Original work: DEKR, HRNet authors
Upstream license: MIT
Upstream source: github.com/HRNet/DEKR
LibreYOLO code: MIT
Weights: No separate artifact grant recorded, distributed by their authors. LibreYOLO does not host or mirror them.
Interpretation: The upstream implementation is MIT. The adapter retrieves the released checkpoint from its source CDN.

Citation

@inproceedings{GengSXZW21,
  title={Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression},
  author={Zigang Geng, Ke Sun, Bin Xiao, Zhaoxiang Zhang, Jingdong Wang},
  booktitle={CVPR},
  year={2021}
}

Copied from the authors' citation block at raw.githubusercontent.com/HRNet/DEKR/main/README.md.