Visual World Knowledge Benchmark for MLLMs
Image-based questions to evaluate atomic world knowledge in multimodal models.
- Visual question answering
- English
- Chinese
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
Image-based questions to evaluate atomic world knowledge in multimodal models.
A large-scale STAIR Captions dataset with 820,310 Japanese image captions for text-to-image and image-to-text generation tasks.
Image-text question-answer pairs in nine languages, including Russian, for evaluating and training multimodal models.
A collection of 38,870 cleaned and verified image-question-answer pairs for training reasoning multimodal models.
A compact set of challenging vision-language tasks with reference answers in four languages.
A multilingual corpus of image captions translated from English into 200 languages by the NLLB-200 model.
Largest combined dataset of plankton photographs with taxonomic labeling.
An English multimodal benchmark for evaluating whether models can write working 3D scene code.
A subset of 500 English question-answer pairs for document scans to evaluate multimodal retrieval and VQA.
A collection of 4,172 scanned engravings and drawings from old books with text captions for image generation and analysis.
Hundreds of thousands of images with detailed texts averaging 200 words and original speech transcripts.
English image and metadata dataset from an ICML 2024 paper for safety fine-tuning of VLMs.
A small set of multimodal feedback for aligning vision-language models.
Set of 3,600 global images with reference descriptions in 36 languages.
Generated metadata and vector representations for 12 million images from the DataComp subset.
An image and text dataset for Handwritten Text Recognition (HTR) tasks in multiple European languages.
300k image-question examples for training vision-language models in LLaMA Factory.
1,000 to 10,000 labeled photos of Korean and English signs to evaluate mobile robot OCR systems.
A training dataset for the ViRA model designed for ultra-high-resolution tasks.
750K CT/MRI images and 7.5M generated VQA examples with expert annotations for training and benchmarking.
204 English questions with 190 images for evaluating long visual search in multimodal agents.
10,013 construction site images with annotations for evaluating scene understanding in vision-language models.
A large Chinese-English corpus based on ALLaVA-4V for training and fine-tuning vision-language models.
English dataset for evaluating models in Visual Question Answering and table comprehension tasks.