Research map / Vision-language and multimodal / Visual question answering
Modulating early visual processing by language
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 |
|---|---|---|---|
| Condition ResNet batch-normalization parameters on the question | ResNet-50 VQA baseline · VQAv1 test-dev, 224x224 | accuracy: 58.05% -> 60.82% | Helped |
| Fine-tune only ResNet stage 4 instead of using the baseline | ResNet-50 VQA baseline · VQAv1 test-dev, 224x224 | accuracy: 58.05% -> 56.91% | Hurt |
| Fine-tune batch-normalization parameters | ResNet-50 VQA baseline · VQAv1 test-dev, 224x224 | accuracy: 58.05% -> 58.98% | Helped |
| Add MODERN modulation to MLB | MLB VQA architecture · VQAv1 test-dev, 448x448 | accuracy: 60.84% -> 63.01% | Helped |
| Condition batch-normalization parameters on the question | GuessWhat?! Oracle crop model · GuessWhat?! Oracle test, Crop | test error: 29.92% -> 25.06% | Helped |
| Add CBN to the model using crop, spatial, and category features | GuessWhat?! Oracle · GuessWhat?! Oracle test, Crop + Spatial + Category | test error: 22.55% -> 19.52% | Helped |
| Add object-crop features to spatial and category features | GuessWhat?! Oracle baseline · GuessWhat?! Oracle test | test error: 21.5% -> 22.55% | Hurt |
| Add CBN to all ResNet stages rather than applying it to none (ablation) | MODERN · VQAv1 validation | accuracy: 56.12% -> 58.56% | Helped |
| Add CBN to all stages rather than only stage 4 (ablation) | MODERN · VQAv1 validation | accuracy: 57.68% -> 58.56% | Helped |
| Add CBN to stage 2 alongside stages 3-4 (ablation) | MODERN · VQAv1 validation | accuracy: 58.29% -> 58.32% | No clear effect |
| Add CBN to stage 1 alongside stages 2-4 (ablation) | MODERN · VQAv1 validation | accuracy: 58.32% -> 58.56% | Helped |
| Add CBN to all ResNet stages rather than applying it to none (ablation) | MODERN · GuessWhat?! Oracle test | test error: 29.92% -> 25.06% | Helped |
Closest papers by meaning
- Language-Instructed Vision Embeddings for Controllable and Generalizable Perception (ICLR 2026 · similarity 0.88)
- MoDA: Modulation Adapter for Fine-Grained Visual Understanding in Instructional MLLMs (ICML 2026 · similarity 0.88)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.88)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.87)
- NeuronEye: Query-Guided Visual Concept Activation for Vision-Language Reasoning (NeurIPS 2026 · similarity 0.87)
- MMRL: Multi-Modal Representation Learning for Vision-Language Models (CVPR 2025 · 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.