OpenPose represents the first real-time multi-person system to jointly detect human body, hand, facial, and foot keypoints (in total 135 keypoints) on single images.
It is a bottom-up approach therefore, it first detects the keypoints belonging to every person in the image, followed by assigning those key-points to a distinct person. In Stage 0, the first 10 layers of the Visual Geometry Group (VGG) are used to create feature maps for the input image. In Stage 1, the top branch predicts a set of 2D confidence maps(S) of body part locations( e.g. elbow, knee etc.). The second branch predicts a set of 2D vector fields (L)ofpartaffinities,whichencodethedegreeofassociation between parts. In Stage 2, the confidence and affinity maps areparsedbygreedyinferencetoproducethe2Dkey-points or joints**.
Divyansh Gupta on a trek of Kedartal #16000 feet
- Functionality:
- 2D real-time multi-person keypoint detection:
- 15 or 18 or 25-keypoint body/foot keypoint estimation. Running time invariant to number of detected people.
- 6-keypoint foot keypoint estimation. Integrated together with the 25-keypoint body/foot keypoint detector.
- 2x21-keypoint hand keypoint estimation. Currently, running time depends on number of detected people.
- 70-keypoint face keypoint estimation. Currently, running time depends on number of detected people.
- 3D real-time single-person keypoint detection:
- 3-D triangulation from multiple single views.
- Synchronization of Flir cameras handled.
- Compatible with Flir/Point Grey cameras, but provided C++ demos to add your custom input.
- Calibration toolbox:
- Easy estimation of distortion, intrinsic, and extrinsic camera parameters.
- Single-person tracking for further speed up or visual smoothing.
- 2D real-time multi-person keypoint detection:
- Input: Image, video, webcam, Flir/Point Grey and IP camera. Included C++ demos to add your custom input.
- Output: Basic image + keypoint display/saving (PNG, JPG, AVI, ...), keypoint saving (JSON, XML, YML, ...), and/or keypoints as array class.
- OS: Ubuntu (14, 16), Windows (8, 10), Mac OSX, Nvidia TX2.
- Training and datasets:
- Others:
- Available: command-line demo, C++ wrapper, and C++ API.
- Python API.
- Unity Plugin.
- CUDA (Nvidia GPU), OpenCL (AMD GPU), and CPU-only (no GPU) versions.
Testing the multi-person body joint estimation with OpenPose
Testing the 3D Reconstruction Module of OpenPose
Authors Gines Hidalgo (left image) and Tomas Simon (right image) testing OpenPose
Inference time comparison between the 3 available pose estimation libraries: OpenPose, Alpha-Pose (fast Pytorch version), and Mask R-CNN:
This analysis was performed using the same images for each algorithm and a batch size of 1. Each analysis was repeated 1000 times and then averaged. This was all performed on a system with a Nvidia 1080 Ti and CUDA 8. Megvii (Face++) and MSRA GitHub repositories were excluded because they only provide pose estimation results given a cropped person. However, they suffer the same problem than Alpha-Pose and Mask R-CNN, their runtimes grow linearly with the number of people.- Features
- Latest Features
- Results
- Installation, Reinstallation and Uninstallation
- Quick Start
- Output
- Speeding Up OpenPose and Benchmark
- Training Code and Foot Dataset
- Send Us Failure Cases and Feedback!
- Citation
- License
Windows portable version: Simply download and use the latest version from the Releases section.
Otherwise, check doc/installation.md for instructions on how to build OpenPose from source.
Most users do not need the OpenPose C++/Python API, but can simply use the OpenPose Demo:
- OpenPose Demo: To easily process images/video/webcam and display/save the results. See doc/demo_overview.md. E.g., run OpenPose in a video with:
# Ubuntu
./build/examples/openpose/openpose.bin --video examples/media/video.avi
:: Windows - Portable Demo
bin\OpenPoseDemo.exe --video examples\media\video.avi
-
Calibration toolbox: To easily calibrate your cameras for 3-D OpenPose or any other stereo vision task. See doc/modules/calibration_module.md.
-
OpenPose C++ API: If you want to read a specific input, and/or add your custom post-processing function, and/or implement your own display/saving, check the C++ API tutorial on examples/tutorial_api_cpp/ and doc/library_introduction.md. You can create your custom code on examples/user_code/ and quickly compile it with CMake when compiling the whole OpenPose project. Quickly add your custom code: See examples/user_code/README.md for further details.
-
OpenPose Python API: Analogously to the C++ API, find the tutorial for the Python API on examples/tutorial_api_python/.
-
Adding an extra module: Check doc/library_add_new_module.md.
-
Standalone face or hand detector:
- Face keypoint detection without body keypoint detection: If you want to speed it up (but also reduce amount of detected faces), check the OpenCV-face-detector approach in doc/standalone_face_or_hand_keypoint_detector.md.
- Use your own face/hand detector: You can use the hand and/or face keypoint detectors with your own face or hand detectors, rather than using the body detector. E.g., useful for camera views at which the hands are visible but not the body (OpenPose detector would fail). See doc/standalone_face_or_hand_keypoint_detector.md.
Output (format, keypoint index ordering, etc.) in doc/output.md.
Check the OpenPose Benchmark as well as some hints to speed up and/or reduce the memory requirements for OpenPose on doc/speed_up_openpose.md.
Please cite these papers in your publications if it helps your research. The body-foot model and any additional functionality (calibration, 3-D reconstruction, etc.) use [Cao et al. 2018]
; the hand and face keypoint detectors use [Cao et al. 2018]
and [Simon et al. 2017]
(the face detector was trained using the same procedure than for hands); and the old (deprecated) body-only model uses [Cao et al. 2017]
.
@inproceedings{cao2018openpose,
author = {Zhe Cao and Gines Hidalgo and Tomas Simon and Shih-En Wei and Yaser Sheikh},
booktitle = {arXiv preprint arXiv:1812.08008},
title = {Open{P}ose: realtime multi-person 2{D} pose estimation using {P}art {A}ffinity {F}ields},
year = {2018}
}
@inproceedings{simon2017hand,
author = {Tomas Simon and Hanbyul Joo and Iain Matthews and Yaser Sheikh},
booktitle = {CVPR},
title = {Hand Keypoint Detection in Single Images using Multiview Bootstrapping},
year = {2017}
}
@inproceedings{cao2017realtime,
author = {Zhe Cao and Tomas Simon and Shih-En Wei and Yaser Sheikh},
booktitle = {CVPR},
title = {Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields},
year = {2017}
}
@inproceedings{wei2016cpm,
author = {Shih-En Wei and Varun Ramakrishna and Takeo Kanade and Yaser Sheikh},
booktitle = {CVPR},
title = {Convolutional pose machines},
year = {2016}
}
Links to the papers:
- OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields
- Hand Keypoint Detection in Single Images using Multiview Bootstrapping
- Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields
- Convolutional Pose Machines
OpenPose is freely available for free non-commercial use, and may be redistributed under these conditions. Please, see the license for further details. Interested in a commercial license? Check this FlintBox link. For commercial queries, use the Directly Contact Organization
section from the FlintBox link and also send a copy of that message to Yaser Sheikh.