Research map / Vision-language and multimodal / Visual question answering
Chain of Reasoning for Visual Question Answering
Chenfei Wu, Jinlai Liu, Xiaojie Wang, Xuan Dong
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 |
|---|---|---|---|
| use chain structure (CoR-3) instead of stacked attention (Stack-3) (ablation) | Mutan attention · VQA 2.0 val | accuracy: 63.90 -> 65.14 | Improved |
| use chain structure instead of parallel attention (Parallel-3) (ablation) | Mutan attention · VQA 2.0 val | accuracy: 63.80 -> 65.14 | Improved |
| use chain structure with MLB attention (CoR-3) instead of plain MLB (ablation) | MLB attention · VQA 2.0 val | accuracy: 62.91 -> 64.96 | Improved |
| use element-wise sum of guided objects for relations instead of RN-style concatenation (ablation) | CoR-2 · VQA 2.0 val | accuracy: 62.46 -> 64.96 | Improved |
| add separate question guidances Gl and Gr for compound and initial objects (ablation) | CoR-2 · VQA 2.0 val | accuracy: 64.73 -> 64.96 | Improved |
| include initial objects at each step's relational reasoning (ablation) | CoR-2 · VQA 2.0 val | accuracy: 64.24 -> 64.96 | Improved |
| refine objects by weighting over previous compound objects (R_ij) rather than R_ji (ablation) | CoR-2 · VQA 2.0 val | accuracy: 64.42 -> 64.96 | Improved |
| increase chain length from 2 to 3 sub-chains (ablation) | CoR · VQA 2.0 val | accuracy: 64.96 -> 65.14 | Improved |
| use CoR-3 versus prior state of the art | CoR-3 · TDIUC | overall accuracy: 85.03 -> 86.91 | Improved |
| use CoR-3 versus prior state of the art | CoR-3 · COCO-QA | accuracy: 66.49 -> 69.38 | Improved |
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.
- Mutan(5) — MUTAN: Multimodal Tucker Fusion for Visual Question Answering (ICCV 2017) · 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
- Visual Chain-of-Thought Prompting for Knowledge-Based Visual Reasoning (AAAI 2024 · similarity 0.84)
- VQA-LOL: Visual Question Answering under the Lens of Logic (ECCV 2020 · similarity 0.84)
- How Transferable Are Reasoning Patterns in VQA? (CVPR 2021 · similarity 0.84)
- Neuro-Symbolic Visual Reasoning: Disentangling "Visual" from "Reasoning" (ICML 2020 · similarity 0.83)
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · similarity 0.83)
- FVQA: Fact-Based Visual Question Answering (TPAMI 2017 · similarity 0.83)
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.
