AudioImageMultimodal
Unify-OmniBench
Unify-OmniBench 统一格式的多模态评测数据集，由 Unify-OmniBench 框架转换生成。 包含四个 benchmark，在 Dataset Viewer 右上角下拉框切换。 数据概览 Config (bench) 题目数 模态 媒体 daily omni ~1197…
1K–10K·Custom / Research-only·Parquet
Datasets
1,204 results
ImageMultimodalText
MathVerse
Dataset Card for MathVerse Dataset Description Paper Information Dataset Examples Leaderboard Citation Dataset Description The capabilities of…
1K–10K·MIT·Parquet
ImageMultimodalText
HoloCount
HoloCount: A Holistic Visual Counting Benchmark for MLLMs Abstract Visual counting is a fundamental pillar of multimodal intelligence,…
1K–10K·CC-BY·Images (folder)
ImageMultimodalText
StepCountQA-SFT
StepCountQA-SFT StepCountQA-SFT is a multimodal supervised fine-tuning (SFT) dataset for visual object counting with step-by-step reasoning. Built…
1M–10M·ODC-BY·Parquet
ImageMultimodalText
MathVerse-lmmseval
Dataset Card for MathVerse This is the version for lmms-eval. This shares the same data with the official…
1K–10K·Parquet
ImageMultimodalText
ImgCode-8.6M
MathCoder-VL: Bridging Vision and Code for Enhanced Multimodal Mathematical Reasoning Repo: Paper: Introduction We introduce MathCoder-VL, a series…
1M–10M·Apache-2.0·Parquet
Image
causalphys
Causal-VL Dataset Causal reasoning VQA dataset with 4 categories × 4 subcategories (3062 questions). Structure Each subcategory contains:…
<1K·Apache-2.0·Images (folder)
ImageMultimodalText
DenseFusion-1M for comprehensive image descriptions
DenseFusion-1M for comprehensive image descriptions
1M–10M·CC-BY·JSON
ImageMultimodalText
Viva Mais Synthetic PT-BR Travel Document Vision Dataset
Viva Mais Synthetic PT-BR Travel Document Vision Dataset
10K–100K·Apache-2.0·JSON
ImageMultimodalText
MathCanvas-Bench
MathCanvas-Bench 🚀 Data Usage from datasets import load dataset dataset = load dataset("shiwk24/MathCanvas-Bench") print(dataset) 📖 Introduction…
1K–10K·Apache-2.0·Parquet
ImageMultimodalText
STRIDE-QA-Dataset
STRIDE-QA Dataset 📦 Dataset STRIDE-QA is a large-scale visual question answering (VQA) dataset for physically grounded spatiotemporal reasoning…
100K–1M·WebDataset