Research map / Vision-language and multimodal / Human and embodied modeling
Teaching Cameras to Feel: Estimating Tactile Physical Properties of Surfaces From Images
Matthew Purri, Kristin Dana
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 |
|---|---|---|---|
| make the regression network cross-modal (ablation) | Single image · SPS single image | avg R2 / %err: 0.34/39.1 -> 0.46/34.3 | Improved |
| add conditional GAN adversarial objective to cross-modal network (ablation) | Single image · SPS single image | avg R2: 0.46 -> 0.49 | Improved |
| add conditional GAN adversarial objective to cross-modal network (ablation) | Single image · SPS single image | avg %err: 34.3 -> 36.2 | Hurt |
| add material class auxiliary classification (ablation) | Cross-modal · SPS single image | avg R2: 0.46 -> 0.45 | Hurt |
| add tactile cluster auxiliary classification (ablation) | Cross-modal · SPS single image | avg R2: 0.46 -> 0.48 | Improved |
| add visuo-tactile cluster auxiliary classification (ablation) | Cross-modal · SPS single image | avg R2: 0.46 -> 0.49 | Improved |
| add adversarial objective on top of visuo-tactile classification (ablation) | Cross-modal + visuo-tactile · SPS single image | avg R2: 0.49 -> 0.50 | Improved |
| use all available images (Late Fusion) instead of a single image | Single image · SPS multi-image | avg R2: 0.50 -> 0.11 | Hurt |
| use equidistant sampling of views instead of single image | Single image · SPS multi-image | avg R2: 0.50 -> 0.51 | Improved |
| use random view sampling instead of single image | Single image · SPS multi-image | avg R2: 0.50 -> 0.47 | Hurt |
| use learned viewpoint selection (NVS) instead of equidistant sampling | Equidistant · SPS multi-image | avg R2: 0.51 -> 0.53 | Improved |
| use variational-bottleneck viewpoint selection (VB-NVS) instead of NVS | NVS · SPS multi-image | avg R2: 0.53 -> 0.55 | Improved |
Closest papers by meaning
- AnyTouch: Learning Unified Static-Dynamic Representation across Multiple Visuo-tactile Sensors (ICLR 2025 · similarity 0.89)
- EgoTac: In-the-wild Tactile Prediction from Egocentric Vision (NeurIPS 2026 · similarity 0.89)
- PhysTacGen: Physics-Aware Visual-Tactile Sensor Image Generation (NeurIPS 2026 · similarity 0.89)
- Sensor-Invariant Tactile Representation (ICLR 2025 · similarity 0.88)
- Touch in the Wild: Learning Fine-Grained Manipulation with a Portable Visuo-Tactile Gripper (NeurIPS 2025 · similarity 0.88)
- Universal Visuo-Tactile Video Understanding for Embodied Interaction (NeurIPS 2025 · similarity 0.87)
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.
