Research map / Point clouds / Shape and surface completion
VisTa3D: A Dataset and Benchmark for Vision, Tactile, and 3D Point Clouds-based Thin Object Reconstruction
Shania Guo, Yeongsik Seo, Andrew Fu, Mei Hao, Iris Xia, Jiwon Lee, Xinyi Xie, Hyoungseob Park, Aaron Dollar, Alex Wong
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 tactile inputs to depth completion (Tactile-DC vs OmniDC) | depth completion · VisTa3D general eval, Default | A1: 0.834 -> 0.844 | Improved |
| add tactile inputs to depth completion (Tactile-DC vs OmniDC) | depth completion · VisTa3D object eval, Default | A1: 0.687 -> 0.719 | Improved |
| add tactile inputs to depth completion (Tactile-DC vs OmniDC) | depth completion · VisTa3D object eval, Linear Fit | A1: 0.342 -> 0.487 | Improved |
| add tactile inputs to depth completion (Tactile-DC vs OmniDC) | depth completion · VisTa3D object eval, Median Scale | A1: 0.439 -> 0.540 | Improved |
| add tactile inputs to depth completion (Tactile-DC vs OmniDC) | depth completion · VisTa3D object eval, Default | AbsRel: 0.274 -> 0.205 | Improved |
| restrict evaluation to thin-object mask instead of all pixels | OmniDC · VisTa3D, Default | A1: 0.834 -> 0.687 | Hurt |
| restrict evaluation to thin-object mask instead of all pixels | Tactile-DC · VisTa3D, Default | A1: 0.844 -> 0.719 | Hurt |
| apply Linear Fit scale alignment instead of raw outputs | DepthAnything V2 · VisTa3D object eval | A1: 0.000 -> 0.058 | Improved |
| apply Linear Fit scale alignment instead of raw outputs | OmniDC · VisTa3D object eval | A1: 0.687 -> 0.342 | Hurt |
| upgrade UniDepth V1 to V2 (after scale correction) | UniDepth · VisTa3D object eval, Linear Fit | A1: 0.049 -> 0.092 | Improved |
| upgrade DepthAnything V1 to V2 (after scale correction) | DepthAnything · VisTa3D object eval, Median Scale | A1: 0.071 -> 0.087 | Improved |
| use multi-view VGGT instead of monocular DepthAnything V2 | depth estimation · VisTa3D object eval, Median Scale | A1: 0.087 -> 0.221 | Improved |
Closest papers by meaning
- 3DReflecNet: A Large-Scale Dataset for 3D Reconstruction of Reflective, Transparent, and Low-Texture Objects (CVPR 2026 · similarity 0.87)
- TouchAnything: Diffusion-Guided 3D Reconstruction from Sparse Robot Touches (ECCV 2026 · similarity 0.87)
- SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images (CVPR 2025 · similarity 0.86)
- Thin-Shell-SfT: Fine-Grained Monocular Non-rigid 3D Surface Tracking with Neural Deformation Fields (CVPR 2025 · similarity 0.85)
- Tactile DreamFusion: Exploiting Tactile Sensing for 3D Generation (NeurIPS 2024 · similarity 0.85)
- Fin3R: Fine-tuning Feed-forward 3D Reconstruction Models via Monocular Knowledge Distillation (NeurIPS 2025 · similarity 0.85)
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.
