Research map / Vision-language and multimodal / Visual question answering
Robust Visual Reasoning via Language Guided Neural Module Networks
Arjun Reddy Akula, Varun Jampani, Soravit Changpinyo, Song-Chun Zhu
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 |
|---|---|---|---|
| replace LG-Conv with standard convolution (ablation) | Ours (with CoSAtt) · CLOSURE | accuracy: 88.0 -> 62.1 | Hurt |
| replace LG-Conv with standard convolution (ablation) | Ours (with CoSAtt) · S-Ref Test | accuracy: 63.3 -> 51.6 | Hurt |
| replace LG-Conv with standard convolution (ablation) | Ours (with CoSAtt) · F-Ref Test | accuracy: 84.3 -> 77.8 | Hurt |
| add image conditioning to the CoSAtt guidance (ablation) | Ours (with CoSAtt) · CLOSURE | accuracy: 80.2 -> 88.0 | Improved |
| add query conditioning to the CoSAtt guidance (ablation) | Ours (with CoSAtt) · CLOSURE | accuracy: 78.8 -> 88.0 | Improved |
| add module argument to the CoSAtt guidance (ablation) | Ours (with CoSAtt) · CLOSURE | accuracy: 82.1 -> 88.0 | Improved |
| add module argument to the CoSAtt guidance (ablation) | Ours (with CoSAtt) · S-Ref Test | accuracy: 61.7 -> 63.3 | Improved |
| add module argument to the CoSAtt guidance (ablation) | Ours (with CoSAtt) · F-Ref Test | accuracy: 80.9 -> 84.3 | Improved |
| use CoSAtt encoder instead of BiSAtt encoder | Ours · CLOSURE | accuracy: 86.1 -> 88.0 | Improved |
| use CoSAtt encoder instead of BiSAtt encoder | Ours · S-Ref Test | accuracy: 59.7 -> 63.3 | Improved |
| use CoSAtt encoder instead of BiSAtt encoder | Ours · CLEVR-Test | accuracy: 99.2 -> 99.2 | No clear effect |
| evaluate on C3-Ref+ novel compositions and contrast sets versus S-Ref | Ours · C3-Ref+ | accuracy | Hurt |
Closest papers by meaning
- Transformer Module Networks for Systematic Generalization in Visual Question Answering (TPAMI 2024 · similarity 0.90)
- Visual Question Reasoning on General Dependency Tree (CVPR 2018 · similarity 0.89)
- Rethinking Token Reduction for Large Vision-Language Models (CVPR 2026 · similarity 0.87)
- From Head to Tail: Towards Balanced Representation in Large Vision-Language Models through Adaptive Data Calibration (CVPR 2025 · similarity 0.87)
- Towards Self-Refinement of Vision-Language Models with Triangular Consistency (NeurIPS 2025 · similarity 0.87)
- Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models (ICLR 2024 · 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.
