Research map / Vision-language and multimodal / Human and embodied modeling
Systematic Visual Reasoning through Object-Centric Relational Abstraction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 slot attention (vs 4x4 grid patches) (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 56.58 -> 85.31 (RMTS); 84.42 -> 92.80 (ID) | Helped |
| add slot attention pre-training (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 58.69 -> 85.31 (RMTS); 53.48 -> 92.80 (ID) | Helped |
| add relational embeddings before transformer (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 74.79 -> 85.31 (RMTS); 81.21 -> 92.80 (ID) | Helped |
| add factorized feature/position embeddings (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 50.26 -> 85.31 (RMTS); 75.40 -> 92.80 (ID) | Helped |
| add relational bottleneck (vs pairwise MLP) (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 63.62 -> 85.31 (RMTS); 76.56 -> 92.80 (ID) | Helped |
| use inner product instead of learned linear bottleneck (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 50.70 -> 85.31 (RMTS); 48.07 -> 92.80 (ID) | Helped |
| use inner product instead of outer product (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 62.84 -> 85.31 (RMTS); 69.58 -> 92.80 (ID) | Helped |
| add transformer for higher-order relations (vs MLP sum) (ablation) | OCRA · ART RMTS/ID m=95 | test accuracy (%): 51.29 -> 85.31 (RMTS); 44.18 -> 92.80 (ID) | Helped |
| OCRA vs GAMR at hardest regime | GAMR · ART same/different m=95 | test accuracy (%): 83.49 -> 87.95 | Helped |
| OCRA vs Slot-Transformer at hardest regime | Slot-Transformer · ART distribution-of-3 m=95 | test accuracy (%): 60.61 -> 86.42 | Helped |
| OCRA vs Slot-Transformer in easiest regime | Slot-Transformer · ART same/different m=0 | test accuracy (%): 98.47 -> 97.31 | No clear effect |
| OCRA vs Slot-Transformer on CLEVR-ART identity rules | Slot-Transformer · CLEVR-ART ID | test accuracy (%): 78.81 -> 77.06 | No clear effect |
Closest papers by meaning
- Slot Abstractors: Toward Scalable Abstract Visual Reasoning (ICML 2024 · similarity 0.92)
- Beyond the Doors of Perception: Vision Transformers Represent Relations Between Objects (NeurIPS 2024 · similarity 0.90)
- ARC Is a Vision Problem! (CVPR 2026 · similarity 0.88)
- What Makes a Maze Look Like a Maze? (ICLR 2025 · similarity 0.88)
- MARVEL: Multidimensional Abstraction and Reasoning through Visual Evaluation and Learning (NeurIPS 2024 · similarity 0.87)
- Mind the GAP: Glimpse-based Active Perception improves generalization and sample efficiency of visual reasoning (ICLR 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.