Research map / Vision-language and multimodal / Visual question answering
MUTAN: Multimodal Tucker Fusion for Visual Question Answering
Hedi Ben-younes, Remi Cadene, Matthieu Cord, Nicolas Thome
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 rank sparsity constraint (R=10) to the Tucker core tensor (ablation) | MUTAN_noR · VQA test-dev | All accuracy: 59.92 -> 60.17 | Improved |
| add rank sparsity constraint (R=10) to the Tucker core tensor (ablation) | MUTAN_noR · VQA val | All accuracy: 57.94 -> 58.16 | Improved |
| replace MLB identity core tensor with a learned full core tensor (MUTAN_noR) | MLB · VQA test-dev | All accuracy: 60.08 -> 59.92 | No clear effect |
| replace MLB with MUTAN fusion (Tucker with rank constraint) | MLB · VQA test-dev | All accuracy: 60.08 -> 60.17 | Improved |
| late fusion of MUTAN with MLB | MUTAN · VQA test-dev | All accuracy: 60.17 -> 61.02 | Improved |
| late fusion of MUTAN with MLB | MUTAN · VQA val | All accuracy: 58.16 -> 58.76 | Improved |
| use MUTAN_noR learned core tensor instead of identity tensor for t from 20 to 220 (ablation) | low-dimensional bilinear fusion · VQA val | accuracy | Improved |
| lower rank R=20 instead of R=60 (ablation) | MUTAN with t_q=t_v=210 · VQA val | accuracy | Improved |
| ensemble of 5 models (with MLB) instead of 3 MUTAN models | MUTAN ensemble · VQA test-dev | All accuracy: 67.03 -> 67.42 | Improved |
| turn off all but one rank projection z_r (ablation) | MUTAN R=20 · VQA val | accuracy per question type | Mixed |
Later papers that compared with this method
6 later accepted papers report MUTAN as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Multi-Modal Answer Validation for Knowledge-Based VQA (AAAI 2022 · better on 1 of 1 shared result)
- Differential Networks for Visual Question Answering (AAAI 2019 · better on 2 of 2 shared results)
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · better on 1 of 1 shared result)
- Deep Attention Neural Tensor Network for Visual Question Answering (ECCV 2018 · better on 6 of 6 shared results)
- Chain of Reasoning for Visual Question Answering (NeurIPS 2018 · better on 2 of 2 shared results)
- Question-Guided Hybrid Convolution for Visual Question Answering (ECCV 2018 · 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
- BLOCK: Bilinear Superdiagonal Fusion for Visual Question Answering and Visual Relationship Detection (AAAI 2019 · similarity 0.91)
- Compact Trilinear Interaction for Visual Question Answering (ICCV 2019 · similarity 0.91)
- Multi-Modal Factorized Bilinear Pooling With Co-Attention Learning for Visual Question Answering (ICCV 2017 · similarity 0.90)
- MUREL: Multimodal Relational Reasoning for Visual Question Answering (CVPR 2019 · similarity 0.89)
- Towards Robust Visual Question Answering via Prompt-Driven Geometric Harmonization (AAAI 2025 · similarity 0.88)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.88)
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.
