Research map / Video / Video generation
MADiff: Motion-Aware Mamba Diffusion Models for Hand Trajectory Prediction on Egocentric Videos
Junyi Ma, Xieyuanli Chen, Wentao Bao, Jingyi Xu, Hesheng Wang
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 |
|---|---|---|---|
| add egomotion features to motion-driven selective scan (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.067 -> 0.065 ADE; 0.113 -> 0.105 FDE | Improved |
| guide denoising with egomotion in MDSS rather than fusing it into fusion module output (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.119 -> 0.065 ADE; 0.156 -> 0.105 FDE | Improved |
| use concatenation instead of summation in MDSS (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.069 -> 0.065 ADE; 0.110 -> 0.105 FDE | Improved |
| use unidirectional scan instead of bidirectional Mamba (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.070 -> 0.065 ADE; 0.109 -> 0.105 FDE | Improved |
| add visual prompt input to waypoints (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.079 -> 0.070 ADE; 0.136 -> 0.112 FDE | Improved |
| add text prompt hand to visual input (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.070 -> 0.065 ADE; 0.112 -> 0.105 FDE | Improved |
| use text prompt hand instead of arm or body (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE: 0.068 -> 0.065 ADE (arm); 0.069 -> 0.065 ADE (body) | Improved |
| use GLIP instead of CLIP backbone (ablation) | MADiff · EgoPAT3D-DT (seen) | ADE/FDE: 0.068 -> 0.065 ADE; 0.108 -> 0.105 FDE | Improved |
| use GLIP instead of ResNet-18 trained from scratch (ablation) | MADiff · EgoPAT3D-DT (unseen) | ADE/FDE: 0.061 -> 0.054 ADE; 0.092 -> 0.086 FDE | Improved |
| use verb-specific manual text prompt for throw (ablation) | MADiff · EK100 verb throw | WDE/FDE: 0.457 -> 0.387 WDE; 0.212 -> 0.180 FDE | Improved |
| use verb-specific manual text prompt for scoop (ablation) | MADiff · EK100 verb scoop | WDE/FDE: 0.436 -> 0.407 WDE; 0.202 -> 0.190 FDE | Improved |
| use automated verb prompt for throw instead of hand (ablation) | MADiff · EK100 verb throw | WDE/FDE: 0.457 -> 0.433 WDE; 0.212 -> 0.201 FDE | Improved |
Closest papers by meaning
- HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing (NeurIPS 2026 · similarity 0.88)
- Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning (ICLR 2025 · similarity 0.88)
- Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data (NeurIPS 2024 · similarity 0.88)
- Whole-Body Conditioned Egocentric Video Prediction (NeurIPS 2025 · similarity 0.87)
- StableHand: Quality-Aware Flow Matching for World-Space Dual-Hand Motion Estimation from Egocentric Video (NeurIPS 2026 · similarity 0.87)
- MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation (NeurIPS 2025 · 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.
