Skip to content

Datasets

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.

Data type

For which models

Find a dataset for your goal

Answered by Gemini 4 · the chat is kept only in this tab

Datasetterryoo/TableVQA-Bench

Table and image visual question answering benchmark

English dataset for evaluating models in Visual Question Answering and table comprehension tasks.

  • Table question answering
  • Visual question answering
  • English
1.1K 351K–10K rows
Datasetjamescalam/unsplash-25k-photos

Unsplash 25K streaming photo dataset

Streaming dataset of 10K to 100K photos from Unsplash servers with English metadata.

  • Image → image
  • Image classification
  • Image → text
  • English
1K 4810K–100K rows
DatasetDeepGlint-AI/DanQing100MTop pick

100M Chinese image-text pairs

A massive 12 TB web corpus from 2024–2025 for pre-training multimodal models.

  • Zero shot image classification
  • Image → text
  • Chinese
1K 5210M–100M rows
DatasetVLR-CVC/DocVQA-2026Top pick

DocVQA 2026 document benchmark

A small document QA benchmark across eight domains for multimodal reasoning.

  • Visual question answering
  • Document question answering
  • Image text → text
  • English
976 74up to 1K rows
Datasetskylenage-ai/DeepVision-103K

Multimodal math problems with images

100k to 1M visually diverse English math problems for training VLM reasoning.

  • Image text → text
  • English
967 37100K–1M rows
DatasetRapidata/OpenAI-4o_t2i_human_preference

Human evaluations for OpenAI 4o image generations

13k images from OpenAI 4o with over 200k real human ratings for training preference models.

  • Text → image
  • Image → text
  • Image classification
  • English
965 3510K–100K rows
Datasetbiglam/britannica-illustrated-pages

Illustrated pages of the Britannica encyclopedia

Over 115k illustrated pages from historical editions of the Britannica encyclopedia (1768–1929).

  • Image → text
  • Image classification
  • English
958 511M–10M rows
Datasetspatialverse/SAGE-3D_VLN_DataTop pick

2 million pairs for Vision-Language Navigation

2 million trajectory-instruction pairs across 1,000 indoor environments for training vision-language navigation.

  • Robotics
  • Image text → text
934 69
DatasetHumanbased-AI/MM-Food-100K

100k food images with metadata

Dish images with metadata for food recognition, classification, and nutritional analysis.

  • Image classification
  • Image → text
  • English
933 84100K–1M rows
Datasetlinxy/LaTeX_OCR

Formula images and LaTeX markup

Pairs of mathematical formula images with LaTeX code for training and evaluating OCR.

  • Image → text
909 195100K–1M rows
DatasetOpenDataArena/MMFineReason-SFT-123K-Qwen3-VL-235B-Thinking

MMFineReason: complex VQA examples for SFT

123k most difficult image-text examples from MMFineReason where Qwen3-VL-4B-Thinking consistently fails.

  • Visual question answering
  • Question answering
  • Text generation
  • English
907 87100K–1M rows
DatasetTeklia/CASIA-HWDB2-line

CASIA handwritten Chinese text base

A line dataset of handwritten Chinese text collected from over a thousand authors.

  • Image → text
  • Chinese
894 2110K–100K rows
Datasetgoogle/docci

DOCCI images with descriptions

Images with detailed human descriptions and pairs of similar frames for evaluating vision models.

  • Text → image
  • Image → text
  • English
889 8210K–100K rows
DatasetOpenGVLab/MMPRTop pick

Large multimodal reasoning dataset for InternVL

Millions of examples for visual logical reasoning training that improved InternVL2.5 and InternVL3 quality.

  • Visual question answering
  • English
883 521M–10M rows
Datasetlaolao77/MMDU

Multi-image multi-turn dialogue benchmark

A set of 110 complex dialogues and over 1600 questions across image series for evaluating multimodal models.

  • Visual question answering
  • Question answering
  • English
863 44up to 1K rows
Datasetneulab/MultiUITop pick

MultiUI multimodal web interface dataset (7.3M)

7.3 million multimodal examples from 1 million websites for training web agents and VLM models.

  • Question answering
  • Visual question answering
  • English
847 531M–10M rows🔒 on request
DatasetChan-Y/pixelart-308k

Synthetic pixel art dataset

A dataset of 308K synthetic pixel art images paired with text descriptions.

  • Text → image
  • Image → text
842 10100K–1M rows
DatasetSakanaAI/JFinBench

JFinBench Japanese financial benchmark

An exam-based benchmark for evaluating LLMs in the Japanese financial domain.

  • Question answering
  • Visual question answering
  • Japanese
836 3up to 1K rows🔒 on request
Datasetvikhyatk/lnqa

LNQA image question answering

A VQA dataset based on Localized Narratives featuring image question-answer pairs.

  • Visual question answering
813 91100K–1M rows
Datasetevayinhao/MIRA-BenchTop pick

VLM spatial reasoning benchmark in mirror scenes

Over 14,000 question-answer pairs on real photos with reflections for evaluating multimodal models.

  • Visual question answering
  • English
803 410K–100K rows🔒 on request
DatasetOleehyO/latex-formulasTop pick

Images and LaTeX formulas for OCR

Large dataset of image-LaTeX pairs for mathematical formula recognition.

  • Image → text
802 1051M–10M rows
DatasetYqjMartin/FLARETop pick

Long video and audio moment retrieval benchmark

Set of 87.7k short video clips with three descriptions each for evaluating multimodal retrieval in long videos.

  • Video text → text
792 2100K–1M rows