Semantics-aware Test-time Adaptation for 3D Human Pose Estimation
Not on the map. Why?
Abstract (arXiv)
This work highlights a semantics misalignment in 3D human pose estimation. For the task of test-time adaptation, the misalignment manifests as overly smoothed and unguided predictions. The smoothing settles predictions towards some average pose. Furthermore, when there are occlusions or truncations, the adaptation becomes fully unguided. To this end, we pioneer the integration of a semantics-aware motion prior for the test-time adaptation of 3D pose estimation. We leverage video understanding and a well-structured motion-text space to adapt the model motion prediction to adhere to video semantics during test time. Additionally, we incorporate a missing 2D pose completion based on the motion-text similarity. The pose completion strengthens the motion prior's guidance for occlusions and truncations. Our method significantly improves state-of-the-art 3D human pose estimation TTA techniques, with more than 12% decrease in PA-MPJPE on 3DPW and 3DHP.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 29 of its 39 references that are in the map. First posted on arXiv 15 Feb 2025.
Builds on
29 of 39 references.
References · 29
- 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
- 2023GPT-4 Technical Reportcited by 18,396
- 2019Expressive Body Capture: 3D Hands, Face, and Body from a Single Imagecited by 1,612
- 2019AMASS: Archive of Motion Capture as Surface Shapescited by 1,236
- 2016Realtime Multi-Person 2D Pose Estimation using Part Affinity Fieldscited by 1,064
- 2017End-to-end Recovery of Human Shape and Posecited by 885
- 2022MotionDiffuse: Text-Driven Human Motion Generation with Diffusion Modelcited by 780
- 2019VIBE: Video Inference for Human Body Pose and Shape Estimationcited by 620
- 2023T2M-GPT: Generating Human Motion from Textual Descriptions with Discrete Representationscited by 588
- 2023MotionGPT: Human Motion as a Foreign Languagecited by 527
- 2016Monocular 3D Human Pose Estimation In The Wild Using Improved CNN Supervisioncited by 494
- 2022MotionCLIP: Exposing Human Motion Generation to CLIP Spacecited by 408
17 more
- 2022TM2T: Stochastic and Tokenized Modeling for the Reciprocal Generation of 3D Human Motions and Textscited by 359
- 2023Humans in 4D: Reconstructing and Tracking Humans with Transformerscited by 350
- 2021HuMoR: 3D Human Motion Model for Robust Pose Estimationcited by 324
- 2021BABEL: Bodies, Action and Behavior with English Labelscited by 265
- 2023WHAM: Reconstructing World-grounded Humans with Accurate 3D Motioncited by 206
- 2021GLAMR: Global Occlusion-Aware Human Mesh Recovery with Dynamic Camerascited by 153
- 2021EgoBody: Human Body Shape and Motion of Interacting People from Head-Mounted Devicescited by 139
- 2021Learning Motion Priors for 4D Human Body Capture in 3D Scenescited by 98
- 2024Expressive Whole-Body 3D Gaussian Avatarcited by 89
- 2023Realistic Full-Body Tracking from Sparse Observations via Joint-Level Modelingcited by 43
- 2021Bilevel Online Adaptation for Out-of-Domain Human Mesh Reconstructioncited by 39
- 2020Inference Stage Optimization for Cross-scenario 3D Human Pose Estimationcited by 35
- 2022PoseTriplet: Co-evolving 3D Human Pose Estimation, Imitation, and Hallucination under Self-supervisioncited by 25
- 2021Out-of-Domain Human Mesh Reconstruction via Dynamic Bilevel Online Adaptationcited by 24
- 2023Cyclic Test-Time Adaptation on Monocular Video for 3D Human Mesh Reconstructioncited by 18
- 2022Domain Adaptive 3D Pose Augmentation for In-the-wild Human Mesh Recoverycited by 12
- 2023On the Calibration of Human Pose Estimationcited by 10
Built on by
Citing papers per year: 2026: 1.
1 paper.