Research map / 3D vision / Human pose estimation
DECO: Dense Estimation of 3D Human-Scene Contact In The Wild
Shashank Tripathi, Agniv Chatterjee, Jean-Claude Passy, Hongwei Yi, Dimitrios Tzionas, Michael J. Black
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 DECO scene+part cross-attention architecture instead of BSTRO | 3D contact estimation · RICH test (RICH-only training) | F1: 0.63 -> 0.70 | Improved |
| add 2D pixel anchoring loss (ablation) | DECO · RICH test | geodesic error (cm): 17.92 -> 10.46 (F1 0.70 -> 0.69) | Improved |
| add scene and part segmentation branch losses (ablation) | DECO · RICH test | geodesic error (cm): 23.54 -> 17.92 | Improved |
| use separate encoders instead of a shared encoder without branch losses (ablation) | DECO · RICH test | geodesic error (cm): 20.85 -> 23.54 | Hurt |
| use HRNet backbone instead of Swin (ablation) | DECO · RICH test | F1: 0.66 -> 0.70 | Improved |
| train on additional PROX and DAMON data | DECO vs BSTRO · DAMON test | F1: 0.46 -> 0.55 (BSTRO vs DECO) | Improved |
| predict contact directly instead of computing geometric contact from reconstructed humans and objects | contact estimation · DAMON subsets | F1: 0.08 -> 0.48 (CHORE vs DECO) | Improved |
| use DECO-inferred contact in PROX body fitting instead of no contact | HPS optimization · PROX quantitative | V2V error: 183.3 -> 171.6 | Improved |
| relax GMoF robust scale rho_C from 5e-02 to 1.0 with DECO contacts (ablation) | HPS optimization · PROX quantitative | V2V error: 171.6 -> 169.0 | Improved |
| increase GMoF robust scale rho_C beyond 1.0 (to 5.0) (ablation) | HPS optimization · PROX quantitative | V2V error: 169.0 -> 183.5 | Hurt |
Closest papers by meaning
- PICO: Reconstructing 3D People In Contact with Objects (CVPR 2025 · similarity 0.91)
- InteractVLM: 3D Interaction Reasoning from 2D Foundational Models (CVPR 2025 · similarity 0.91)
- ContactField: Implicit Field Representation for Multi-Person Interaction Geometry (NeurIPS 2024 · similarity 0.89)
- Pose Priors from Language Models (CVPR 2025 · similarity 0.88)
- Learning Explicit Contact for Implicit Reconstruction of Hand-Held Objects from Monocular Images (AAAI 2024 · similarity 0.88)
- Learning Dense Hand Contact Estimation from Imbalanced Data (NeurIPS 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.
