Research map / Video / Video segmentation and tracking
Look Ma, No Hands! Agent-Environment Factorization of Egocentric Videos
Matthew Chang, Aditya Prakash, Saurabh Gupta
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 |
|---|---|---|---|
| Use past-frame attention (VIDM) instead of single-frame LDM inpainting | inpainting model · EPIC held-out inpainting | PSNR: 28.29 -> 32.26 | Improved |
| Use past-frame attention instead of single-frame LDM | inpainting model · EPIC held-out inpainting | FID: 27.29 -> 10.37 | Improved |
| Fine-tune single-frame LDM on EPIC images | Latent Diffusion · EPIC held-out inpainting | PSNR: 28.29 -> 28.27 | No clear effect |
| Use VIDM with 3 frames history instead of DLFormer | inpainting model · EPIC held-out inpainting | runtime: 106.4s / image -> ~13.6s / image | Improved |
| Black out the agent mask instead of raw image | Mask R-CNN · VISOR active objects, Mask R-CNN | AR_all@5: 0.263 -> 0.236 | Hurt |
| Inpaint agent with single-frame LDM | Mask R-CNN · VISOR active objects, Mask R-CNN | AR_all@5: 0.263 -> 0.259 | Mixed |
| Inpaint agent with VIDM (no factorization) | Mask R-CNN · VISOR active objects, Mask R-CNN | AR_all@1: 0.137 -> 0.163 | Improved |
| Use both raw and VIDM-inpainted images (factorization) (ablation) | Mask R-CNN · VISOR active objects, Mask R-CNN | AR_all@5: 0.268 -> 0.379 | Improved |
| Add ground-truth environment image as input | 3D hand-held object reconstruction · ObMan | F1@5mm: 0.41 -> 0.48 | Improved |
| Add ground-truth environment image as input | 3D hand-held object reconstruction · ObMan | F1@10mm: 0.62 -> 0.68 | Improved |
| Add ground-truth environment image as input | 3D hand-held object reconstruction · ObMan | chamfer distance: 1.23 -> 1.06 | Improved |
| Inpaint hands with VIDM instead of masking | affordance model · Region of Interaction 0% slack | Average Precision: 47.50 -> 50.77 | 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.
- DLFormer — DLFormer: Discrete Latent Transformer for Video Inpainting (CVPR 2022) · 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
- The Surprising Effectiveness of Video Diffusion Models for Hand Motion Reconstruction (NeurIPS 2026 · similarity 0.88)
- Egocentric Procedure Parsing (ECCV 2026 · similarity 0.87)
- Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning (ICLR 2025 · similarity 0.87)
- Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data (NeurIPS 2024 · similarity 0.86)
- HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing (NeurIPS 2026 · similarity 0.86)
- ComPose: When to Trust Hands for Object Pose Tracking (NeurIPS 2026 · similarity 0.86)
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.
