Research map / Vision-language and multimodal / Human and embodied modeling
Geoclidean: Few-Shot Generalization in Euclidean Geometry
Joy Hsu, Jiajun Wu, Noah Goodman
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 |
|---|---|---|---|
| human vs pretrained vision features | ImageNet-pretrained VGG16/RN50/InV3/ViT · Geoclidean 37 concepts | average accuracy: 0.91 -> 0.70 | Hurt |
| high-level instead of low-level features (ablation) | ViT · Geoclidean 37 concepts | average accuracy: 0.55 -> 0.70 | Improved |
| high-level instead of low-level features (ablation) | VGG16 · Geoclidean 37 concepts | average accuracy: 0.55 -> 0.60 | Improved |
| high-level instead of low-level features (ablation) | ResNet50 · Geoclidean 37 concepts | average accuracy: 0.55 -> 0.53 | Hurt |
| high-level instead of low-level features (ablation) | InceptionV3 · Geoclidean 37 concepts | average accuracy: 0.54 -> 0.60 | Improved |
| ViT high-level vs convolutional high-level features | ViT · Geoclidean 37 concepts | average accuracy: 0.60 -> 0.70 | Improved |
| Far negatives instead of Close negatives | humans · Geoclidean concept llc | accuracy: 0.6767 -> 0.9233 | Improved |
| Far negatives instead of Close negatives | humans · Geoclidean concept ccc | accuracy: 0.6667 -> 0.8767 | Improved |
| Far instead of Close negatives | humans · Geoclidean rhomboid | accuracy: 0.9700 -> 0.9300 | Hurt |
| ViT high-level vs humans on circle concepts | ViT · Geoclidean concept ccc | accuracy: 0.77 -> 0.90 | Improved |
| ViT high-level vs humans on cccc | ViT · Geoclidean concept cccc | accuracy: 0.85 -> 0.95 | Improved |
| InceptionV3 high-level alignment with humans | InceptionV3 · Geoclidean 74 tasks | Pearson correlation: 0.2575 | Improved |
Closest papers by meaning
- Alignment of CNN and Human Judgments of Geometric and Topological Concepts (AAAI 2025 · similarity 0.88)
- GeoX: Geometric Problem Solving Through Unified Formalized Vision-Language Pre-training (ICLR 2025 · similarity 0.88)
- Do Large Language Models Truly Understand Geometric Structures? (ICLR 2025 · similarity 0.87)
- GeoBench: Rethinking Multimodal Geometric Problem-Solving via Hierarchical Evaluation (ICLR 2026 · similarity 0.87)
- GeoTikzBridge: Advancing Multimodal Code Generation for Geometric Perception and Reasoning (CVPR 2026 · similarity 0.86)
- GIQ: Benchmarking 3D Geometric Reasoning of Vision Foundation Models with Simulated and Real Polyhedra (ICLR 2026 · similarity 0.86)
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.
