Research map / Vision-language and multimodal / Visual question answering
Multi-Modality Latent Interaction Network for Visual Question Answering
Peng Gao, Haoxuan You, Zhanpeng Zhang, Xiaogang Wang, Hongsheng Li
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 |
|---|---|---|---|
| stack 8 MLI blocks instead of 1 (ablation) | MLIN · VQA v2.0 val | accuracy: 66.04 -> 66.53 | Improved |
| use 6 by 6 summarization vectors instead of 3 by 3 (ablation) | MLIN · VQA v2.0 val | accuracy: 65.63 -> 66.04 | Improved |
| use 12 by 12 summarization vectors instead of 6 by 6 (ablation) | MLIN · VQA v2.0 val | accuracy: 66.04 -> 66.21 | Improved |
| use element-wise product instead of addition as interaction operator (ablation) | MLIN · VQA v2.0 val | accuracy: 65.69 -> 66.04 | Improved |
| use MUTAN bilinear fusion instead of product as interaction operator (ablation) | MLIN · VQA v2.0 val | accuracy: 66.04 -> 66.20 | Improved |
| use product instead of concatenation as interaction operator (ablation) | MLIN · VQA v2.0 val | accuracy: 65.99 -> 66.04 | No clear effect |
| use embedding dimension 1024 instead of 512 (ablation) | MLIN · VQA v2.0 val | accuracy: 66.04 -> 66.18 | Improved |
| use linear propagation instead of self attention (ablation) | MLIN · VQA v2.0 val | accuracy: 65.84 -> 66.04 | Improved |
| use linear propagation instead of dual attention (ablation) | MLIN · VQA v2.0 val | accuracy: 66.01 -> 66.04 | No clear effect |
| use key-query feature gathering instead of transpose attention (ablation) | MLIN · VQA v2.0 val | accuracy: 65.78 -> 66.04 | Improved |
| use 16 parallel heads instead of 8 (ablation) | MLIN · VQA v2.0 val | accuracy: 65.84 -> 66.19 | Improved |
| finetune BERT at lr 1/10 instead of freezing (ablation) | MLI-8 + BERT · VQA v2.0 val | accuracy: 65.51 -> 67.83 | Improved |
Closest papers by meaning
- MRA-Net: Improving VQA Via Multi-Modal Relation Attention Network (TPAMI 2020 · similarity 0.92)
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.91)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.91)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.90)
- Assessing Modality Bias in Video Question Answering Benchmarks with Multimodal Large Language Models (AAAI 2025 · similarity 0.89)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · 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.
