Research map / Vision-language and multimodal / Visual question answering
Multimodal Residual Learning for Visual QA
Read the paperCheck 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 |
|---|---|---|---|
| add residual learning with shortcuts (MRN vs MN) at L=3 (ablation) | MRN · VQA test-dev OE | accuracy (All): 59.87 -> 61.68 | Helped |
| compare MRN against MN at best depth for each (ablation) | MRN · VQA test-dev OE | accuracy (All): 60.90 -> 61.68 | Helped |
| number of learning blocks L=1 to 4 (ablation) | MRN · VQA test-dev OE | accuracy (All): 58.85 -> 59.44 -> 60.53 -> 60.42 | Mixed |
| add extra embedding for question vector (c) (ablation) | MRN (b) · VQA test-dev OE | accuracy (All): 60.53 -> 60.19 | Hurt |
| use identity shortcuts (d) instead of linear mapping shortcuts (ablation) | MRN (b) · VQA test-dev OE | accuracy (All): 60.53 -> 59.69 | Hurt |
| add extra question-only shortcut in multiple blocks (e) (ablation) | MRN (b) · VQA test-dev OE | accuracy (All): 60.53 -> 60.20 | No clear effect |
| model (b) versus (a) (ablation) | MRN · VQA test-dev OE | accuracy (All): 60.17 -> 60.53 | Helped |
| replace VGG-19 with ResNet-152 visual features (1k answers) (ablation) | MRN · VQA test-dev OE | accuracy (All): 60.53 -> 61.45 | Helped |
| increase target answers from 1k to 2k (ResNet features) (ablation) | MRN · VQA test-dev OE | accuracy (All): 61.45 -> 61.68 | Helped |
| increase target answers from 2k to 3k (ResNet features) (ablation) | MRN · VQA test-dev OE | accuracy (All): 61.68 -> 61.47 | Hurt |
| initialize question GRU with Skip-Thought Vectors (ablation) | MRN (a) · VQA test-dev OE | accuracy (All): 58.97 -> 59.38 | Helped |
| add Bayesian Dropout (ablation) | MRN (a) · VQA test-dev OE | accuracy (All): 59.38 -> 59.74 | Helped |
Closest papers by meaning
- Multimodal Commonsense Knowledge Distillation for Visual Question Answering (Student Abstract) (AAAI 2025 · similarity 0.86)
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
- Object Attribute Matters in Visual Question Answering (AAAI 2024 · similarity 0.86)
- Rethinking Token Reduction for Large Vision-Language Models (CVPR 2026 · similarity 0.85)
- Question Aware Vision Transformer for Multimodal Reasoning (CVPR 2024 · similarity 0.85)
- Predictive Regularization Against Visual Representation Degradation in Multimodal Large Language Models (CVPR 2026 · similarity 0.85)
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.