Research map / Point clouds / Shape and surface completion
TouchDream: 3D Object Completion through Imagined Touch
Yuanbo Wang, Xinning Wang, Zhaoxuan Zhang, Changlong Wang, Qianchen Xia, Xiaopeng Wei, Xin Yang
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 generated touch guidance instead of symmetry guidance (ablation) | SymmCompletion-style refinement · PCN | L1 CD avg (x10^3): 6.28 -> 6.05 | Improved |
| combine symmetry guidance with touch guidance (ablation) | TouchDream · PCN | L1 CD avg (x10^3): 6.05 -> 6.02 | No clear effect |
| use simulated ground truth touch instead of generated touch (ablation) | TouchDream · PCN | L1 CD avg (x10^3): 6.05 -> 5.93 | Improved |
| use skeleton points from ground truth instead of generated touch (ablation) | TouchDream · PCN | L1 CD avg (x10^3): 6.05 -> 6.04 | No clear effect |
| add test-time pose noise (R/T 0.02/0.01) without augmentation (ablation) | TouchDream · PCN | L1 CD avg (x10^3): 6.053 -> 6.183 | Hurt |
| add pose-noise data augmentation during training (ablation) | TouchDream · PCN with pose noise R/T 0.02/0.01 | L1 CD avg (x10^3): 6.183 -> 6.087 | Improved |
| add imagined touch guidance to coarse completion | SymmCompletion · ShapeNet-34 unseen categories | L2 CD avg (x10^3): 0.97 -> 0.66 | Improved |
| add imagined touch guidance to coarse completion | SymmCompletion · KITTI (trained on ShapeNet Car) | FD / MMD: FD 2.54 -> 1.43; MMD 1.72 -> 0.71 | Improved |
| generate a touch latent per pose with diffusion | TouchDream · PCN evaluation cost | evaluation time: 6.7 hours for 1,200 shapes | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- SymmCompletion — SymmCompletion: High-Fidelity and High-Consistency Point Cloud Completion with Symmetry Guidance (AAAI 2025) · better on 6 of 6 shared results
- AnchorFormer — AnchorFormer: Point Cloud Completion From Discriminative Nodes (CVPR 2023) · better on 2 of 2 shared results
- PointSea — PointSea: Point Cloud Completion via Self-structure Augmentation (IJCV 2025) · better on 1 of 1 shared result
- CRA-PCN — CRA-PCN: Point Cloud Completion with Intra- and Inter-level Cross-Resolution Transformers (AAAI 2024) · better on 1 of 1 shared result
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
- TouchAnything: Diffusion-Guided 3D Reconstruction from Sparse Robot Touches (ECCV 2026 · similarity 0.91)
- Tactile DreamFusion: Exploiting Tactile Sensing for 3D Generation (NeurIPS 2024 · similarity 0.89)
- Touch2Shape: Touch-Conditioned 3D Diffusion for Shape Exploration and Reconstruction (CVPR 2025 · similarity 0.89)
- Tactile-Augmented Radiance Fields (CVPR 2024 · similarity 0.87)
- PhysTacGen: Physics-Aware Visual-Tactile Sensor Image Generation (NeurIPS 2026 · similarity 0.85)
- Tactile MNIST: Benchmarking Active Tactile Perception (NeurIPS 2026 · 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.
