Difference between revisions of "Pose Estimation"
(→Add MoveNet) |
|||
| (9 intermediate revisions by the same user not shown) | |||
| Line 12: | Line 12: | ||
! Sr No !! Model !! Developer !! Key Points !! Source !! License | ! Sr No !! Model !! Developer !! Key Points !! Source !! License | ||
|- | |- | ||
| − | | 1 || | + | | 1 || MediaPipe || Google || Tracking 33 key points on the human body, offering cross-platform, customizable ML solutions || [https://github.com/google/mediapipe MediaPipe GitHub] || Apache 2.0 |
|- | |- | ||
| − | | 2 || MoveNet || Google Research || Detecting 17 critical key points of the human body || [https://github.com/tensorflow/tfjs-models/tree/master/posenet MoveNet GitHub] || Apache 2.0 | + | | 2 || Detectron2 || Facebook AI Research || High-performance codebase for object detection and segmentation, including pose estimation || [https://github.com/facebookresearch/detectron2 Detectron2 GitHub] || Apache 2.0 |
| + | |- | ||
| + | | 3 || OpenPose || Carnegie Mellon University || Detecting key points of the human body, including hand, facial, and foot || [https://github.com/CMU-Perceptual-Computing-Lab/openpose OpenPose GitHub] || MIT | ||
| + | |- | ||
| + | | 4 || MoveNet || Google Research || Detecting 17 critical key points of the human body || [https://github.com/tensorflow/tfjs-models/tree/master/posenet MoveNet GitHub] || Apache 2.0 | ||
| + | |- | ||
| + | | 5 || PoseNet || Google Research || Detecting different body parts, providing comprehensive skeletal information || [https://github.com/tensorflow/tfjs-models/tree/master/posenet PoseNet GitHub] || Apache 2.0 | ||
| + | |- | ||
| + | | 6 || DCPose || Deep Dual Consecutive Network || Detecting human pose from multiple frames, addressing motion blur and occlusions || [https://github.com/DeepDualConsecutivePose/dcpose DCPose GitHub] || MIT | ||
| + | |- | ||
| + | | 7 || DensePose || Facebook AI Research || Mapping human-based pixels from an RGB image to the 3D surface of a human body || [https://github.com/facebookresearch/DensePose DensePose GitHub] || Apache 2.0 | ||
| + | |- | ||
| + | | 8 || HigherHRNet || HRNet || Addressing scaling differences in pose prediction, especially for shorter people || [https://github.com/HRNet/HigherHRNet HigherHRNet GitHub] || MIT | ||
| + | |- | ||
| + | | 9 || Lightweight OpenPose || Daniil-Osokin || Real-time inference with minimal accuracy drop, detecting human poses through key points || [https://github.com/Daniil-Osokin/lightweight-human-pose-estimation Lightweight OpenPose GitHub] || MIT | ||
| + | |- | ||
| + | | 10 || AlphaPose || MVIG-SJTU || Detecting multiple individuals in various scenes, achieving high mAP on COCO and MPII datasets || [https://github.com/MVIG-SJTU/AlphaPose AlphaPose GitHub] || MIT | ||
|} | |} | ||
| + | |||
| + | |||
| + | [[File:Pose detection overlay.gif|720px|thumb|Mediapipe Pose Detection]] | ||
| + | |||
| + | <gallery> | ||
| + | Pose_example1.png|About to Stand | ||
| + | Pose_example2.png|Standing but error in leg detection | ||
| + | Pose_example3.png|Foreground missed | ||
| + | Pose_example4.png|Hoodie | ||
| + | </gallery> | ||
| + | |||
| + | [[File:Poseoutput white orig.gif|720px|thumb|Mediapipe Pose Detection]] | ||
Latest revision as of 06:47, 7 June 2024
Pose estimation is a computer vision technique that predicts the configuration of a person's or object's joints or parts in an image or video.
It involves detecting and tracking the position and orientation of these parts, usually represented as keypoints.
Pose estimation is widely used in applications such as motion capture, human-computer interaction, augmented reality, and robotics. The process typically involves training machine learning models on large datasets of annotated images to accurately identify and locate the keypoints.
Pose Estimation Related Models[edit]
| Sr No | Model | Developer | Key Points | Source | License |
|---|---|---|---|---|---|
| 1 | MediaPipe | Tracking 33 key points on the human body, offering cross-platform, customizable ML solutions | MediaPipe GitHub | Apache 2.0 | |
| 2 | Detectron2 | Facebook AI Research | High-performance codebase for object detection and segmentation, including pose estimation | Detectron2 GitHub | Apache 2.0 |
| 3 | OpenPose | Carnegie Mellon University | Detecting key points of the human body, including hand, facial, and foot | OpenPose GitHub | MIT |
| 4 | MoveNet | Google Research | Detecting 17 critical key points of the human body | MoveNet GitHub | Apache 2.0 |
| 5 | PoseNet | Google Research | Detecting different body parts, providing comprehensive skeletal information | PoseNet GitHub | Apache 2.0 |
| 6 | DCPose | Deep Dual Consecutive Network | Detecting human pose from multiple frames, addressing motion blur and occlusions | DCPose GitHub | MIT |
| 7 | DensePose | Facebook AI Research | Mapping human-based pixels from an RGB image to the 3D surface of a human body | DensePose GitHub | Apache 2.0 |
| 8 | HigherHRNet | HRNet | Addressing scaling differences in pose prediction, especially for shorter people | HigherHRNet GitHub | MIT |
| 9 | Lightweight OpenPose | Daniil-Osokin | Real-time inference with minimal accuracy drop, detecting human poses through key points | Lightweight OpenPose GitHub | MIT |
| 10 | AlphaPose | MVIG-SJTU | Detecting multiple individuals in various scenes, achieving high mAP on COCO and MPII datasets | AlphaPose GitHub | MIT |
About to Stand
Standing but error in leg detection
Foreground missed
Hoodie
