mlsanityData to 14 Oct 2026

PoseSyn: Synthesizing Diverse 3D Pose Data from In-the-Wild 2D Data

arXiv 2503.13025 ↗First posted 17 Mar 2025cs.CV cs.AIcited by 1

Not on the map. Why?

Abstract (arXiv)

Despite considerable efforts to enhance the generalization of 3D pose estimators without costly 3D annotations, existing data augmentation methods struggle in real world scenarios with diverse human appearances and complex poses. We propose PoseSyn, a novel data synthesis framework that transforms abundant in the wild 2D pose dataset into diverse 3D pose image pairs. PoseSyn comprises two key components: Error Extraction Module (EEM), which identifies challenging poses from the 2D pose datasets, and Motion Synthesis Module (MSM), which synthesizes motion sequences around the challenging poses. Then, by generating realistic 3D training data via a human animation model aligned with challenging poses and appearances PoseSyn boosts the accuracy of various 3D pose estimators by up to 14% across real world benchmarks including various backgrounds and occlusions, challenging poses, and multi view scenarios. Extensive experiments further confirm that PoseSyn is a scalable and effective approach for improving generalization without relying on expensive 3D annotations, regardless of the pose estimator's model size or design.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 56 of its 76 references that are in the map. First posted on arXiv 17 Mar 2025.

Builds on

56 of 76 references.

References · 56

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  3. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  4. 2022Generative Adversarial Networkscited by 5,627
  5. 2023Adding Conditional Control to Text-to-Image Diffusion Modelscited by 5,503
  6. 2021LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairscited by 1,555
  7. 2016Realtime Multi-Person 2D Pose Estimation using Part Affinity Fieldscited by 1,064
  8. 2017End-to-end Recovery of Human Shape and Posecited by 885
  9. 2019Learning to Reconstruct 3D Human Pose and Shape via Model-fitting in the Loopcited by 627
  10. 2023T2M-GPT: Generating Human Motion from Textual Descriptions with Discrete Representationscited by 588
  11. 2018Simple Baselines for Human Pose Estimation and Trackingcited by 502
  12. 2016Monocular 3D Human Pose Estimation In The Wild Using Improved CNN Supervisioncited by 494
38 more
  1. 2017AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actionscited by 449
  2. 2023Humans in 4D: Reconstructing and Tracking Humans with Transformerscited by 350
  3. 2022Is synthetic data from generative models ready for image recognition?cited by 320
  4. 2018Learning 3D Human Dynamics from Videocited by 316
  5. 2020HybrIK: A Hybrid Analytical-Neural Inverse Kinematics Solution for 3D Human Pose and Shape Estimationcited by 307
  6. 2024Champ: Controllable and Consistent Human Image Animation with 3D Parametric Guidancecited by 272
  7. 2022CLIFF: Carrying Location Information in Full Frames into Human Pose and Shape Estimationcited by 238
  8. 2021A-NeRF: Articulated Neural Radiance Fields for Learning Human Shape, Appearance, and Posecited by 233
  9. 2023WHAM: Reconstructing World-grounded Humans with Accurate 3D Motioncited by 206
  10. 2017Single-Shot Multi-Person 3D Pose Estimation From Monocular RGBcited by 195
  11. 2016Panoptic Studio: A Massively Multiview System for Social Interaction Capturecited by 173
  12. 2022Fake it till you make it: Learning transferable representations from synthetic ImageNet clonescited by 172
  13. 2023DatasetDM: Synthesizing Data with Perception Annotations Using Diffusion Modelscited by 123
  14. 2022HuMMan: Multi-Modal 4D Human Dataset for Versatile Sensing and Modelingcited by 117
  15. 2023Dataset Diffusion: Diffusion-based Synthetic Dataset Generation for Pixel-Level Semantic Segmentationcited by 112
  16. 2024Multi-HMR: Multi-Person Whole-Body Human Mesh Recovery in a Single Shotcited by 101
  17. 2023EMDB: The Electromagnetic Database of Global 3D Human Pose and Shape in the Wildcited by 97
  18. 2021PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimationcited by 89
  19. 2017AI Challenger : A Large-scale Dataset for Going Deeper in Image Understandingcited by 73
  20. 2023Freestyle Layout-to-Image Synthesiscited by 67
  21. 2021Learning to Estimate Robust 3D Human Mesh from In-the-Wild Crowded Scenescited by 65
  22. 2023FreeMask: Synthetic Images with Dense Annotations Make Stronger Segmentation Modelscited by 61
  23. 2016Synthesizing Training Images for Boosting Human 3D Pose Estimationcited by 52
  24. 2023DGInStyle: Domain-Generalizable Semantic Segmentation with Image Diffusion Models and Stylized Semantic Controlcited by 45
  25. 2018Jointly Optimize Data Augmentation and Network Training: Adversarial Data Augmentation in Human Pose Estimationcited by 44
  26. 2021When Human Pose Estimation Meets Robustness: Adversarial Algorithms and Benchmarkscited by 41
  27. 2016MoCap-guided Data Augmentation for 3D Pose Estimation in the Wildcited by 37
  28. 2017Adversarial PoseNet: A Structure-aware Convolutional Network for Human Pose Estimationcited by 35
  29. 2021AdaptPose: Cross-Dataset Adaptation for 3D Human Pose Estimation by Learnable Motion Generationcited by 32
  30. 2023Reliability in Semantic Segmentation: Can We Use Synthetic Data?cited by 24
  31. 2024Dataset Enhancement with Instance-Level Augmentationscited by 23
  32. 2020Adversarial Semantic Data Augmentation for Human Pose Estimationcited by 22
  33. 2022PLIKS: A Pseudo-Linear Inverse Kinematic Solver for 3D Human Body Estimationcited by 18
  34. 2024DiverGen: Improving Instance Segmentation by Learning Wider Data Distribution with More Diverse Generative Datacited by 15
  35. 2022PoseTrans: A Simple Yet Effective Pose Transformation Augmentation for Human Pose Estimationcited by 14
  36. 2023From Fake to Real: Pretraining on Balanced Synthetic Images to Prevent Spurious Correlations in Image Recognitioncited by 12
  37. 2022Multi-Person 3D Pose and Shape Estimation via Inverse Kinematics and Refinementcited by 11
  38. 2024Data Augmentation for Image Classification using Generative AIcited by 11

50 most cited of 56 All 56 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026ExpertEdit: Learning Skill-Aware Motion Editing from Expert Videoscited by 1