Research map / Vision-language and multimodal / Human and embodied modeling
REX: Reasoning-Aware and Grounded Explanation
Shi Chen, Qi Zhao
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 |
|---|---|---|---|
| Explicitly model word-to-region correspondence with adaptive grounding | VisualBert explanation generation · GQA-REX val | Grounding: 33.52 -> 67.95 | Improved |
| Use reasoning-aware, grounded explanations instead of VQAE explanations | VisualBert · GQA-REX val | BLEU-4: 42.56 -> 54.59 | Improved |
| Use reasoning-aware, grounded explanations instead of VQAE explanations | VisualBert · GQA-val | accuracy: 65.19 -> 66.16 | Improved |
| Use reasoning-aware, grounded explanations instead of EXP explanations | VisualBert transfer learning · GQA balanced val with 1%, 5%, and 10% answer annotations | accuracy: 51.32 -> 57.07; 56.34 -> 61.28; 57.65 -> 61.90 | Improved |
| Use reasoning-aware, grounded explanations instead of EXP explanations | VisualBert transfer learning · GQA-OOD val with 1%, 5%, and 10% answer annotations | accuracy: 35.26 -> 40.03; 41.20 -> 45.02; 43.15 -> 45.98 | Improved |
| Train on explanations as well as answers rather than answers alone | VisualBert under multi-task learning · GQA balanced val, 1%, 5%, and 10% answer annotations | accuracy: 41.41 -> 41.70; 48.53 -> 49.36; 51.79 -> 52.83 | Mixed |
| Increase visual-attribute recognition | VisualBert-REX · GQA reasoning analysis | answer correctness correlation: Color 0.742; Material 0.708; Relation 0.182 | Improved |
Closest papers by meaning
- Rex-Thinker: Grounded Object Referring via Chain-of-Thought Reasoning (ICLR 2026 · similarity 0.89)
- The Abstraction Gap in Vision-Language Causal Reasoning (ICML 2026 · similarity 0.88)
- No Labels, No Problem: Training Visual Reasoners with Multimodal Verifiers (ICLR 2026 · similarity 0.87)
- CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning (CVPR 2026 · similarity 0.87)
- CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization (CVPR 2026 · similarity 0.87)
- ReXTime: A Benchmark Suite for Reasoning-Across-Time in Videos (NeurIPS 2024 · 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.
