Research map / Vision-language and multimodal / Visual question answering
RMLVQA: A Margin Loss Approach for Visual Question Answering With Language Biases
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 |
|---|---|---|---|
| Replace AdaVQA cosine margins with RMLVQA's adaptive angular-margin approach and full components | UpDn VQA model · VQA-CP v2 test | Overall accuracy: 54.02 -> 60.41 | Helped |
| Replace AdaVQA cosine margins with RMLVQA's adaptive angular-margin approach and full components | UpDn VQA model · VQA v2 validation | Overall accuracy: 46.98 -> 59.99 | Helped |
| Add angular margins instead of cosine margins (ablation) | RMLVQA-Base · VQA-CP v2 test | Overall accuracy: AdaVQA 54.02; RMLVQA-Base 57.24 | Helped |
| Add angular margins instead of cosine margins (ablation) | RMLVQA-Base · VQA v2 validation | Overall accuracy: AdaVQA 46.98; RMLVQA-Base 49.72 | Helped |
| Add Gaussian randomization to frequency-based margins (ablation) | RMLVQA-Base · VQA-CP v2 test | Overall accuracy: 57.24 -> 57.97 | Helped |
| Add Gaussian randomization to frequency-based margins (ablation) | RMLVQA-Base · VQA v2 validation | Overall accuracy: 49.72 -> 56.84 | Helped |
| Add the bias-injecting component (ablation) | RMLVQA with randomized margins · VQA-CP v2 test | Overall accuracy: 57.97 -> 59.44 | Helped |
| Add the bias-injecting component (ablation) | RMLVQA with randomized margins · VQA v2 validation | Overall accuracy: 56.84 -> 60.49 | Helped |
| Add instance-specific learnable margins (ablation) | RMLVQA with bias injection · VQA-CP v2 test | Overall accuracy: 59.44 -> 59.87 | Helped |
| Add instance-specific learnable margins (ablation) | RMLVQA with bias injection · VQA v2 validation | Overall accuracy: 60.49 -> 59.60 | Hurt |
| Add supervised contrastive loss (ablation) | RMLVQA with learnable margins · VQA-CP v2 test | Overall accuracy: 59.87 -> 60.41 | Helped |
| Add supervised contrastive loss (ablation) | RMLVQA with learnable margins · VQA v2 validation | Overall accuracy: 59.60 -> 59.99 | Helped |
Closest papers by meaning
- Language‑Bias‑Resilient Visual Question Answering via Adaptive Multi‑Margin Collaborative Debiasing (NeurIPS 2025 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.91)
- Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models (ICLR 2024 · similarity 0.90)
- NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples (NeurIPS 2024 · similarity 0.90)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · similarity 0.90)
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.