Research map / 3D vision / Human pose estimation
Ego-Human Motion Prediction with 3D-Aware LLM
Yujin Bae, Jaewoo Jeong, HYEONSEONG KIM, KUK-JIN YOON
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| jointly decode pose and language in a single pass with 3D scene and GRPO (Ego3DLM vs UniEgoMotion) | Ego3DLM · Nymeria future motion prediction (3 modes) | JPE: 424.3 -> 364.5 | Improved |
| Ego3DLM vs UniEgoMotion on motion tracking | Ego3DLM · Nymeria motion tracking | APE: 152.2 -> 96.4 | Improved |
| single-pass joint decoding vs separate decoding (EgoLM) | Ego3DLM · Nymeria past motion narration | Bleu-4: 0.0649 -> 0.1107 | Improved |
| add 3D scene input (ablation) | Ego3DLM SFT · Nymeria motion tracking | APE: 105.8 -> 98.3 | Improved |
| add semantic QA pretraining (ablation) | Ego3DLM SFT · Nymeria motion tracking | APE: 122.1 -> 98.3 | Improved |
| add spatial QA pretraining (ablation) | Ego3DLM SFT · Nymeria future motion prediction (3 modes) | JPE: 378.3 -> 368.3 | Improved |
| add spatial scene reasoning (SSR) step before outputs (ablation) | Ego3DLM SFT · Nymeria future motion description | R@3: 0.2638 -> 0.2706 | Improved |
| add inter-modal matching reward on top of motion+text rewards (ablation) | Ego3DLM GRPO · Nymeria future motion prediction (3 modes) | JPE: 369.6 -> 364.5 | No clear effect |
| add inter-modal matching reward (ablation) | Ego3DLM GRPO · Nymeria future motion description | R@3: 0.2726 -> 0.2911 | Improved |
| GRPO finetuning over SFT only (ablation) | Ego3DLM · Nymeria motion tracking | APE: 98.3 -> 96.4 | Improved |
| add motion reward alone vs SFT only (ablation) | Ego3DLM GRPO · Nymeria future motion prediction (3 modes) | FID: 0.0190 -> 0.0149 | Improved |
| add 3D scene to baselines (ablation) | UniEgoMotion · Nymeria future motion prediction (3 modes) | JPE: 439.5 -> 424.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UniEgoMotion — UniEgoMotion: A Unified Model for Egocentric Motion Reconstruction, Forecasting, and Generation (ICCV 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multimodal Sense-Informed Forecasting of 3D Human Motions (CVPR 2024 · similarity 0.91)
- Gaze Beyond the Frame: Forecasting Egocentric 3D Visual Span (NeurIPS 2025 · similarity 0.90)
- MolmoMotion: Forecasting Point Trajectories in 3D with Language Instruction (NeurIPS 2026 · similarity 0.90)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.88)
- Vision-Guided Action: Enhancing 3D Human Motion Prediction with Gaze-informed Affordance in 3D Scenes (CVPR 2025 · similarity 0.88)
- GEM: A Generalizable Ego-Vision Multimodal World Model for Fine-Grained Ego-Motion, Object Dynamics, and Scene Composition Control (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
