Human Pose Estimation
Locating body, hand and face keypoints of people in images and video.
Tier Grades
Tier A: Frontier AI labs
feasible
Big tech releases human-centric foundation models.
Tier B: Indonesian AI labs
feasible
Training pose models of 100M parameters is within an Indonesian lab reach.
Tier C: Institution units such as ITB
feasible
A 4-author team produced ViTPose, a size a university unit can match.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Released pose models support worker safety and ergonomics proofs of concept.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeViTPose scales from 100M to 1B parameters; training compute is not stated in the abstract.
- DataNot in the sources read.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 4 authors on the ViTPose paper (arXiv 2204.12484).
- Time horizonYear.
- Shelf-life signalsHuman-centric foundation models (for example Sapiens) replace task models, On-device pose reaches desktop accuracy
- Pipeline roleTier A and academia publish pose models; Tier C adapts them to local settings; Tier D uses them for safety and activity monitoring pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
RTMPosevisionexemplar
ViTPosevisionexemplar
BlazePose GHUMvisionexemplar
OpenPosevisionexemplar
Sapiens2visionproduced bySapiensvisionproduced byRTMWvisionproduced byRTMOvisionproduced by
PoseAnythingvisionproduced by
ED-Posevisionproduced byYOLO-Posevisionproduced by
HRFormervisionproduced by
SimCCvisionproduced by
ViPNASvisionproduced by
Lite-HRNetvisionproduced by
DEKRvisionproduced by
InterNet (InterHand2.6M)visionproduced byMediaPipe Handsvisionproduced by
VoxelPosevisionproduced byRSNvisionproduced by
HigherHRNetvisionproduced byHRNetvisionproduced byHRNet (pose)visionproduced by
MSPNvisionproduced by+16
VideoPose3Dvisionproduced bySimpleBaseline (pose)visionproduced byDensePosevisionproduced by
AlphaPosevisionproduced byKeypoint R-CNNvisionproduced by
Associative Embeddingvisionproduced byStacked Hourglassvisionproduced byCPM (Convolutional Pose Machines)visionproduced byDeepPosevisionproduced byYOLOX-Posevisionproduced byMoveNetvisionproduced byPoseNetvisionproduced byBodyPixvisionproduced byMediaPipe vision modelsvisionproduced by
Intel human pose estimationvisionproduced by
NVIDIA TAO human understanding modelsvisionproduced byLinks
R&D TopicHuman Pose Estimation
ModelsOpenPoseCarnegie Mellon University, visionRTMPoseOpenMMLab, visionViTPoseUniversity of Sydney, visionBlazePose GHUMGoogle, vision+36 more
OpenPoseCarnegie Mellon University, visionRTMPoseOpenMMLab, visionViTPoseUniversity of Sydney, visionBlazePose GHUMGoogle, vision+36 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 40
MediaPipe vision modelsGoogle, visionproduced byHRNetUniversity of Science and Technology of China (USTC), Microsoft Research, visionproduced byHRFormerPeking University, Microsoft Research, visionproduced byLite-HRNetHuazhong University of Science and Technology, Microsoft Research, visionproduced bySapiensMeta, visionproduced bySapiens2Meta, visionproduced byBlazePose GHUMGoogle, visionexemplarHigherHRNetUniversity of Illinois Urbana-Champaign, Microsoft, visionproduced byHRNet (pose)University of Science and Technology of China (USTC), Microsoft Research, visionproduced byInterNet (InterHand2.6M)Seoul National University, Meta, visionproduced byMediaPipe HandsGoogle, visionproduced byViPNASThe Chinese University of Hong Kong, SenseTime, visionproduced by+28
AlphaPoseShanghai Jiao Tong University, visionproduced byAssociative EmbeddingUniversity of Michigan, Tsinghua University, visionproduced byBodyPixGoogle, visionproduced byDEKRUniversity of Science and Technology of China (USTC), Microsoft Research, visionproduced byKeypoint R-CNNMeta, visionproduced byMoveNetGoogle, visionproduced byNVIDIA TAO human understanding modelsNVIDIA, visionproduced byPoseNetGoogle, visionproduced byRSNTsinghua University, Megvii, visionproduced byVideoPose3DETH Zurich, Meta, visionproduced byViTPoseUniversity of Sydney, JD.com, visionexemplarVoxelPoseMicrosoft Research, University of Science and Technology of China (USTC), visionproduced byYOLOX-PoseOpenMMLab, visionproduced byCPM (Convolutional Pose Machines)Carnegie Mellon University, visionproduced byDeepPoseGoogle, visionproduced byDensePoseMeta, visionproduced byED-PoseIDEA Research (International Digital Economy Academy), visionproduced byIntel human pose estimationIntel, visionproduced byMSPNMegvii, visionproduced byOpenPoseCarnegie Mellon University, visionexemplarPoseAnythingTel Aviv University, visionproduced byRTMOOpenMMLab, visionproduced byRTMPoseOpenMMLab, visionexemplarRTMWOpenMMLab, visionproduced bySimCCTsinghua University, visionproduced bySimpleBaseline (pose)Microsoft Research, visionproduced byStacked HourglassUniversity of Michigan, visionproduced byYOLO-Posevisionproduced byPapers
- OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity FieldsZhe Cao et al., arXiv preprint, 2018 founding
- Sapiens: Foundation for Human Vision ModelsRawal Khirodkar et al., ECCV 2024 (per arXiv comment), 2024 state of the art
- RTMPose: Real-Time Multi-Person Pose Estimation based on MMPoseTao Jiang et al., arXiv preprint, 2023 state of the art
- ViTPose: Simple Vision Transformer Baselines for Human Pose EstimationYufei Xu et al., arXiv preprint, 2022 state of the art
- Deep High-Resolution Representation Learning for Human Pose EstimationKe Sun et al., CVPR2019 (per arXiv comment), 2019 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2204.12484 https://arxiv.org/abs/2204.12484
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Human Pose Estimation", https://aurion.semai.id/topics/human-pose-estimation/, records of October 2026.