Research map / Vision-language and multimodal / Human and embodied modeling
One-Shot Recognition of Any Material Anywhere Using Contrastive Learning with Physics-Based Rendering
Manuel S. Drehwald, Sagi Eppel, Jolina Li, Han Hao, Alan Aspuru-Guzik
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 data augmentation during MatSim training (ablation) | MatSim+M · MatSim Set2 | top-1 accuracy: 0.61 -> 0.91 | Improved |
| add transparent-vessel scenes to training data (ablation) | MatSim+M · MatSim Set1 Subclass | top-1 accuracy: 0.72 -> 0.78 | Improved |
| add gradual material mixtures to training data (ablation) | MatSim+M · MatSim Set2 | top-1 accuracy: 0.72 -> 0.91 | Improved |
| use semi-hard loss instead of hard loss (ablation) | MatSim training · MatSim benchmark | top-1 accuracy | Improved |
| crop material region as input to CLIP (ablation) | Open CLIP H14 · MatSim Set2 | top-1 accuracy: 0.47 -> 0.77 | Improved |
| mask background to black for CLIP (ablation) | Open CLIP H14 · MatSim Set1 All | top-1 accuracy: 0.44 -> 0.40 | Hurt |
| add cropping and masking to MatSim net (ablation) | MatSim · MatSim Set2 | top-1 accuracy: 0.73 -> 0.91 | Improved |
| train on MatSim instead of using CLIP on semantic matching | MatSim+M vs Open CLIP H14+C · FMD | matching accuracy: 0.82 -> 0.66 | Hurt |
| train on MatSim instead of human-similarity net | Human Similarity+C · MatSim Set1 Subclass | top-1 accuracy: 0.60 -> 0.78 | Improved |
Closest papers by meaning
- Infusing Synthetic Data with Real-World Patterns for Zero-Shot Material State Segmentation (NeurIPS 2024 · similarity 0.91)
- Hierarchical Material Recognition from Local Appearance (ICCV 2025 · similarity 0.89)
- Harnessing the Power of Foundation Models for Accurate Material Classification (CVPR 2026 · similarity 0.88)
- MaRI: Material Retrieval Integration across Domains (CVPR 2025 · similarity 0.85)
- Material Palette: Extraction of Materials from a Single Image (CVPR 2024 · similarity 0.85)
- Intrinsic Image Diffusion for Indoor Single-view Material Estimation (CVPR 2024 · similarity 0.83)
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.
