Research map / Vision-language and multimodal / Visual question answering
Linguistically Routing Capsule Network for Out-of-Distribution Visual Question Answering
Qingxing Cao, Wentao Wan, Keze Wang, Xiaodan Liang, Liang Lin
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 linguistically routing (unary + binary + parser) to a regular network (ablation) | FiLM-based baseline · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 78.19 -> 85.58 | Improved |
| add linguistic routing (ablation) | MCAN-based baseline · VQA-CP v2 test | accuracy: 51.15 -> 52.18 | Improved |
| add unary routing only (binary potential 0) (ablation) | no-routing baseline · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 78.19 -> 82.17 | Improved |
| add binary potential 1 on top of unary routing (ablation) | LR-Capsule with unary only · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 82.17 -> 85.58 | Improved |
| raise binary potential from 1 to 2 (ablation) | LR-Capsule · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 85.58 -> 79.16 | Hurt |
| raise binary potential from 0 to 0.5 (ablation) | LR-Capsule · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 82.17 -> 82.25 | No clear effect |
| add parser-guided binary potential instead of learned binary potential (ablation) | LR-Capsule with unary+binary · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 78.29 -> 85.58 | Improved |
| keep softmax-normalized single unary value instead of removing normalization (+scale) (ablation) | LR-Capsule · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 81.73 -> 85.58 | Improved |
| use single unary value instead of elementwise multiplicative vector (+mul) (ablation) | LR-Capsule · CLEVR-CoGenT val (trained on condition A) condition B | accuracy: 78.01 -> 82.17 | Improved |
| add LR-Capsule to RUBi on VQA-CP v2 | RUBi · VQA-CP v2 test | accuracy: 47.11 -> 52.19 | Improved |
| add linguistic routing to MCAN on VQA v2 val | MCAN · VQA v2 val | accuracy: 67.2 -> 67.04 | No clear effect |
| replace LMH ensemble with LR-Capsule on Number questions | LMH · VQA-CP v2 test | accuracy (Number): 44.46 -> 28.37 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MuRel — MUREL: Multimodal Relational Reasoning for Visual Question Answering (CVPR 2019) · better on 1 of 1 shared result
- RUBi — RUBi: Reducing Unimodal Biases for Visual Question Answering (NeurIPS 2019) · better on 1 of 1 shared result
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
- Dynamic Capsule Attention for Visual Question Answering (AAAI 2019 · similarity 0.91)
- Found a Reason for me? Weakly-supervised Grounded Visual Question Answering using Capsules (CVPR 2021 · similarity 0.90)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.90)
- Graph-Structured Representations for Visual Question Answering (CVPR 2017 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.89)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · 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.
