Research map / Vision-language and multimodal / Visual question answering
Compositional Substitutivity of Visual Reasoning for Visual Question Answering
Chuanhao Li, Zhen Li, Chenchen Jing, Yuwei Wu, Mingliang Zhai, Yunde Jia
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 SUPS support-set training framework | MAC · GQA-SPS Word SPS | Consistency: 47.5 -> 55.6 | Improved |
| add SUPS to five VQA baselines | ViLT, mPLUG, BEiT-3, MAC, LXMERT · GQA-SPS HM consistency | HM Consistency: LXMERT 65.5 -> 67.3; BEiT-3 65.3 -> 66.2 | Improved |
| add SUPS to baselines on in-distribution GQA | BEiT-3 · GQA test-dev | Accuracy: 60.8 -> 61.8 | Improved |
| add SUPS to MAC on in-distribution GQA | MAC · GQA test-dev | Accuracy: 53.1 -> 53.5 | No clear effect |
| add SUPS to LXMERT on VQA-Rephrasings | LXMERT · VQA-Rephrasings | CS(4): 60.03 -> 60.93 | Improved |
| use support sets for plain data augmentation instead of reconstruction (ablation) | LXMERT · GQA-SPS | Consistency (Word / Visual Entity / Referent): Visual Entity 73.30 -> 69.83 (DA question), Referent 56.63 -> 55.62 | Hurt |
| build support sets from same-batch samples (OSB) instead of back-translation and retrieval (ablation) | LXMERT · GQA-SPS | Consistency: SUPS Word 71.43 vs OSB 69.57 | Hurt |
| use both question and image support sets (ablation) | LXMERT+SUPS · GQA-SPS | Consistency: Referent 58.26 -> 58.39 | Improved |
| increase sampled support number K above 10 (ablation) | LXMERT+SUPS · GQA-SPS | HM Consistency | Hurt |
| scale up to large vision-language models in zero-shot | CogVLM, LLaVA and others · GQA-SPS | HM Consistency: CogVLM 57.7 vs LXMERT 65.5 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BEiT-3 — BEiT: BERT Pre-Training of Image Transformers (ICLR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Maintaining Reasoning Consistency in Compositional Visual Question Answering (CVPR 2022 · similarity 0.93)
- Core-to-Global Reasoning for Compositional Visual Question Answering (AAAI 2025 · similarity 0.93)
- GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question Answering (CVPR 2019 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
- Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs? (ICML 2025 · similarity 0.91)
- Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models (ICLR 2025 · similarity 0.91)
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.
