Research map / Vision-language and multimodal / Robot action and navigation
EgoMotion: Hierarchical Vision-Language Learning and Diffusion for Egocentric Motion Generation
Ruibing Hou, Mingyue Zhou, Yuwei Gui, Mingshuang Luo, Bingpeng Ma, Hong Chang, Shiguang Shan, Xilin Chen
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 |
|---|---|---|---|
| replace VQ-VAE with RVQ-VAE (ablation) | autoregressive generator · Nymeria Ego-VL motion generation | FID: 0.051 -> 0.028 | Improved |
| replace VQ-VAE with RVQ-VAE (ablation) | autoregressive generator · Nymeria Ego-VL motion generation | Jerk: 0.0041 -> 0.0023 | Improved |
| replace VQ-VAE with RVQ-VAE (ablation) | masked generator · Nymeria Ego-VL motion generation | R@Top1: 62.7 -> 69.1 | Improved |
| use diffusion instead of autoregressive generator (ablation) | EgoMotion · Nymeria Ego-VL motion generation | FID: 0.028 -> 0.015 | Improved |
| use diffusion instead of masked generator (ablation) | EgoMotion · Nymeria Ego-VL motion generation | FID: 0.031 -> 0.015 | Improved |
| move diffusion from raw space to VAE latent space (ablation) | EgoMotion · Nymeria Ego-VL motion generation | Foot Sliding: 1.56 -> 1.05 | Improved |
| move diffusion from raw space to VAE latent space (ablation) | EgoMotion · Nymeria Ego-VL motion generation | Inference time (ms): 295 -> 226 | Improved |
| move diffusion from raw space to VAE latent space (ablation) | EgoMotion · Nymeria Ego-VL motion generation | FID: 0.015 -> 0.018 | Hurt |
| two-stage training instead of freezing the VLM (ablation) | EgoMotion · Nymeria Ego-VL motion generation | FID: 0.030 -> 0.015 | Improved |
| two-stage training instead of joint tuning (ablation) | EgoMotion · Nymeria Ego-VL motion generation | FID: 0.027 -> 0.015 | Improved |
| two-stage training instead of joint tuning (ablation) | EgoMotion · Nymeria Ego-VL motion generation | Training time (h): 202.6 -> 112.4 | Improved |
| two-stage training instead of joint tuning (ablation) | EgoMotion · Nymeria Ego-VL motion generation | R@Top1: 71.9 -> 69.8 | Hurt |
Closest papers by meaning
- EgoDyn-Bench: Evaluating Ego-Motion Understanding in Vision-Centric Foundation Models for Autonomous Driving (ECCV 2026 · similarity 0.91)
- EgoGen: An Egocentric Synthetic Data Generator (CVPR 2024 · similarity 0.89)
- EgoLM: Multi-Modal Language Model of Egocentric Motions (CVPR 2025 · similarity 0.89)
- EgoProx: Evaluating MLLMs on Egocentric 3D Proximity Reasoning Across a Cognitive Hierarchy (CVPR 2026 · similarity 0.89)
- EgoThinker: Unveiling Egocentric Reasoning with Spatio-Temporal CoT (NeurIPS 2025 · similarity 0.88)
- EARL: Towards a Unified Analysis-Guided Reinforcement Learning Framework for Egocentric Interaction Reasoning and Pixel Grounding (ICML 2026 · similarity 0.87)
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.
