Research map / Vision-language and multimodal / Visual question answering
CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning
Check 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 |
|---|---|---|---|
| Use spatial attention (CNN+LSTM+SA) rather than CNN+LSTM on attribute-query questions | VQA models on CLEVR · CLEVR attribute queries | accuracy | Helped |
| Require models to compare two object attributes | VQA models on CLEVR · CLEVR attribute comparison | accuracy: approximately 50% | Hurt |
| Use CNN image features for counting | CNN+LSTM and CNN+LSTM+MCB · CLEVR counting | accuracy | No clear effect |
| Add spatial attention | CNN+LSTM on CLEVR counting · CLEVR counting | accuracy: 52% | Helped |
| Use same-attribute relations instead of spatial relations | CNN+LSTM+SA · CLEVR query questions | accuracy: 93% vs 78% | Hurt |
| Use tree topology rather than chain topology | CNN+LSTM+SA · CLEVR query questions | accuracy: 92% vs 74% | Hurt |
| Use tree topology rather than chain topology | VQA models on CLEVR count questions · CLEVR count questions | accuracy: 55% vs 49% | Hurt |
| Increase effective question size | VQA models on CLEVR · CLEVR query questions | accuracy | Hurt |
| Require nontrivial relative spatial reasoning rather than allowing absolute-position shortcuts | VQA models on CLEVR · CLEVR spatial questions | accuracy | Hurt |
| Swap the color palettes associated with cubes and cylinders at test time | CNN+LSTM+SA trained on Condition A · CLEVR query color, cubes/cylinders | accuracy: 85% to 51% | Hurt |
| Swap the color palettes associated with cubes and cylinders at test time | CNN+LSTM+SA trained on Condition A · CLEVR query material, cubes/cylinders | accuracy: 90% vs 81% | Hurt |
Closest papers by meaning
- ChartMuseum: Testing Visual Reasoning Capabilities of Large Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- Long Grounded Thoughts: Synthesizing Visual Problems and Reasoning Chains at Scale (ICML 2026 · similarity 0.90)
- Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models (ICML 2025 · similarity 0.89)
- ConMe: Rethinking Evaluation of Compositional Reasoning for Modern VLMs (NeurIPS 2024 · similarity 0.89)
- VisRes Bench: On Evaluating the Visual Reasoning Capabilities of VLMs (CVPR 2026 · similarity 0.89)
- PRISM-Bench: A Benchmark of Puzzle-Based Visual Tasks with CoT Error Detection (NeurIPS 2026 · similarity 0.89)
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.