Research map / Video / Action recognition
EPIC Fields: Marrying 3D Geometry and Video Understanding
Vadim Tschernezki, Ahmad Darkhalil, Zhifan Zhu, David Fouhey, Iro Laina, Diane Larlus, Dima Damen, Andrea Vedaldi
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 overlap-based homography filtering instead of uniform frame sampling (ablation) | EPIC Fields camera reconstruction · 30-video reconstruction comparison | Successful reconstructions: 27/30 -> 30/30 | Improved |
| Use overlap-based homography filtering instead of uniform frame sampling (ablation) | EPIC Fields camera reconstruction · 30-video reconstruction comparison | Avg. #3D Points: 23,142 -> 27,763 | Improved |
| Use overlap-based homography filtering instead of uniform frame sampling (ablation) | EPIC Fields camera reconstruction · 30-video reconstruction comparison | Avg. Reprojection Error: 0.836 -> 0.798 | Improved |
| Use overlap-based homography filtering instead of uniform frame sampling (ablation) | EPIC Fields camera reconstruction · 30-video reconstruction comparison | Avg. Registration Rate: 89.0% -> 98.6% | Improved |
| Represent reference objects in 3D and reproject masks using camera poses rather than keeping pixels fixed in 2D (ablation) | Fixed-mask VOS baseline · VISOR VOS Static | J&F: 17.8 -> 48.4 | Improved |
| Represent reference objects in 3D and reproject masks using camera poses rather than keeping pixels fixed in 2D (ablation) | Fixed-mask VOS baseline · VISOR VOS SS + Dyn | J&F: 12.0 -> 29.6 | Improved |
| Use 3D neural rendering rather than 2D motion grouping | Unsupervised object segmentation · EPIC Fields UDOS SS | mAP: 12.78 -> 25.55 | Improved |
| Use 2D optical-flow-based Motion Grouping rather than NeuralDiff | Unsupervised object segmentation · EPIC Fields UDOS Dynamic | mAP: 55.58 -> 64.27 | Improved |
| Evaluate frames during actions with substantial object motion rather than easier out-of-action frames | Neural rendering baselines · EPIC Fields D-NVS | PSNR: 22.14 -> 19.88 -> 18.36 | Hurt |
Closest papers by meaning
- HD-EPIC: A Highly-Detailed Egocentric Video Dataset (CVPR 2025 · similarity 0.87)
- Layered Motion Fusion: Lifting Motion Segmentation to 3D in Egocentric Videos (CVPR 2025 · similarity 0.87)
- Probing the 3D Awareness of Visual Foundation Models (CVPR 2024 · similarity 0.87)
- From an Image to a Scene: Learning to Imagine the World from a Million 360° Videos (NeurIPS 2024 · similarity 0.86)
- GEOPHYS: The Geometry of Physical Plausibility (NeurIPS 2026 · similarity 0.86)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · 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.
