Research map / Vision-language and multimodal / Cross-modal learning
Multi-modal Data Spectrum: Multi-modal Datasets are Multi-dimensional
Divyam Madaan, Varshan Muhunthan, Kyunghyun Cho, Sumit Chopra
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 |
|---|---|---|---|
| scale Cambrian-1 from smaller models to 34B (ablation) | Cambrian-1 · MMMU | overall accuracy and modality reliance | Mixed |
| scale Cambrian-1 model size (ablation) | Cambrian-1 · MMBench | image-only dependency | Mixed |
| scale Cambrian-1 model size (ablation) | Cambrian-1 · AI2D | text-only dependency | Hurt |
| scale Cambrian-1 model size (ablation) | Cambrian-1 · POPE | accuracy | No clear effect |
| use newer MLLMs (LLaVA-Next, Qwen2.5-VL, Qwen3-VL instead of Cambrian-1 8B) (ablation) | MLLM modality dependence · GQA, MMBench, POPE, MMMU-Pro | intra-modality bias | No clear effect |
| switch from Cambrian-1 to Qwen models (ablation) | MLLM · MMBench | accuracy and image-only accuracy: around ten percent accuracy gain | Improved |
| design benchmarks to remove text-only bias | VQA benchmark design · MMBench and other newer benchmarks | image-only dependency: image-only +41% over random on MMBench | Hurt |
| keep answers differing for the same image with yes/no question pairs | POPE, MME · POPE, MME | single-modality performance | Improved |
| add a 'None of the above' abstention option | GPT-5 and Gemini 2.5 Pro · irrelevant image or question cases | abstention / reliance on uni-modal cues | No clear effect |
Closest papers by meaning
- Evaluating Cross-Modal Reasoning Ability and Problem Charactaristics with Multimodal Item Response Theory (ICLR 2026 · similarity 0.92)
- Assessing Modality Bias in Video Question Answering Benchmarks with Multimodal Large Language Models (AAAI 2025 · similarity 0.91)
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.90)
- Understanding Information Storage and Transfer in Multi-Modal Large Language Models (NeurIPS 2024 · similarity 0.89)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 2026 · 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.
