datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
VQAv2VQAonline
VQAonline
🌐 Homepage | 🤗 Dataset | 📖 arXiv
Dataset Description
We introduce VQAonline, the first VQA dataset in which all contents originate from an authentic use case.
VQAonline includes 64K visual questions sourced from an online question answering community (i.e., StackExchange).
It differs from prior datasets; examples include that it contains:
(1) authentic context that clarifies the question
(2) an answer the individual asking the question validated as… See the full description on the dataset page: https://huggingface.co/datasets/ChongyanChen/VQAonline.viet-cultural-vqaVietnamese Cultural VQA Dataset is a comprehensive multimodal dataset focusing on Vietnamese cultural heritage.
It contains 28,505 images across 12 cultural categories with 119,012 question-answer pairs in Vietnamese and English.
The dataset covers diverse aspects of Vietnamese culture including architecture, cuisine, traditional clothing,
landscapes, festivals, folk culture, traditional games, sports, handicrafts, musical instruments, daily life,
and transportation.Kvasir-VQA-x1
Kvasir-VQA-x1
A Multimodal Dataset for Medical Reasoning and Robust MedVQA in Gastrointestinal Endoscopy
Kvasir-VQA-x1 on GitHub |
Original Image from Kvasir-VQA(Simula Datasets) |
Paper
🔗 MediaEval Medico 2025 Challenge uses this dataset. We encourage you to check out and participate!
Overview
Kvasir-VQA-x1 is a large-scale dataset designed to benchmark medical visual question answering (MedVQA) in gastrointestinal (GI) endoscopy. It introduces 159,549 new QA… See the full description on the dataset page: https://huggingface.co/datasets/SimulaMet/Kvasir-VQA-x1.IllusionChar_train
IllusionChar — Training Set
Dataset summary
This repository contains the training split of IllusionChar, the optical character recognition (OCR) component of Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. The task is to transcribe a hidden, case-sensitive alphanumeric sequence from an illusory image, or return No illusion when no sequence is embedded.
Sequences contain 3–5 characters drawn from digits, uppercase Latin letters, and… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/IllusionChar_train.VizWiz-VQA
Dataset Card for "VizWiz-VQA"
Large-scale Multi-modality Models Evaluation Suite
Accelerating the development of large-scale multi-modality models (LMMs) with lmms-eval
🏠 Homepage | 📚 Documentation | 🤗 Huggingface Datasets
This Dataset
This is a formatted version of VizWiz-VQA. It is used in our lmms-eval pipeline to allow for one-click evaluations of large multi-modality models.
@inproceedings{gurari2018vizwiz,
title={Vizwiz grand… See the full description on the dataset page: https://huggingface.co/datasets/lmms-lab-encoder/VizWiz-VQA.vqa-rad
Dataset Card for VQA-RAD
Dataset Description
VQA-RAD is a dataset of question-answer pairs on radiology images. The dataset is intended to be used for training and testing
Medical Visual Question Answering (VQA) systems. The dataset includes both open-ended questions and binary "yes/no" questions.
The dataset is built from MedPix, which is a free open-access online database of medical images.
The question-answer pairs were manually generated by a team of clinicians.… See the full description on the dataset page: https://huggingface.co/datasets/flaviagiammarino/vqa-rad.IllusionChar_test
IllusionChar — Test Set
Dataset summary
This repository contains the public test split of IllusionChar, the OCR benchmark introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. Each metadata row can be paired across source-condition, illusion, filtered-illusion, illusionless-control, and filtered-illusionless-control conditions.
The expected output for an illusion-bearing or source-condition image is an exact, case-sensitive… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/IllusionChar_test.path-vqa
Dataset Card for PathVQA
Dataset Description
PathVQA is a dataset of question-answer pairs on pathology images. The dataset is intended to be used for training and testing
Medical Visual Question Answering (VQA) systems. The dataset includes both open-ended questions and binary "yes/no" questions.
The dataset is built from two publicly-available pathology textbooks: "Textbook of Pathology" and "Basic Pathology", and a
publicly-available digital library: "Pathology… See the full description on the dataset page: https://huggingface.co/datasets/flaviagiammarino/path-vqa.OK-VQAMNIST_train
IllusionMNIST — Training Set
Dataset summary
This repository contains the training split of IllusionMNIST, introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. The dataset is intended for training models to recognize MNIST digits embedded as visual illusions (pareidolia) in generated scenes and to reject images that contain no illusion.
MNIST source-condition images were sampled and resized to 512 × 512 pixels, combined… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/MNIST_train.FashionMnist_train
IllusionFashionMNIST — Training Set
Dataset summary
This repository contains the training split of IllusionFashionMNIST, one of the four datasets introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. It is designed to train and evaluate models on the recognition of Fashion-MNIST categories embedded as visual illusions (pareidolia) in generated scenes.
The source-condition images are sampled from Fashion-MNIST and resized to… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/FashionMnist_train.ZwZ-RL-VQA
ZwZ-RL-VQA: Region-to-Image Distilled Training Data for Fine-Grained Perception
This synthetic dataset is generated via Region-to-Image Distillation (R2I) for training multimodal large language models (MLLMs) on fine-grained perception tasks without test-time tool use.
📖 Overview
The Zooming without Zooming (ZwZ) method transforms "zooming" from an inference-time tool into a training-time primitive:
Zoom-in Synthesis: Strong teacher models (Qwen3-VL-235B, GLM-4.5V)… See the full description on the dataset page: https://huggingface.co/datasets/inclusionAI/ZwZ-RL-VQA.IllusionAnimals_train
IllusionAnimals — Training Set
Dataset summary
This repository contains the training split of IllusionAnimals, one of the four benchmarks introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. It supports training models to identify animal categories embedded as visual illusions (pareidolia) in generated scenes and to recognize when no illusion is present.
The source-condition animal images were generated with SDXL-Lightning.… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/IllusionAnimals_train.OCR-VQA
Dataset Card for "OCR-VQA"
More Information needed
FashionMnist_test
IllusionFashionMNIST — Test Set
Dataset summary
This repository contains the public test split of IllusionFashionMNIST, introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. Each metadata row identifies a Fashion-MNIST target and can be paired across five image conditions: source-condition, illusion, filtered illusion, illusionless control, and filtered illusionless control.
The source-condition images originate from… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/FashionMnist_test.Awesome_Spatial_VQA_BenchmarksMNIST_test
IllusionMNIST — Test Set
Dataset summary
This repository contains the public test split of IllusionMNIST, introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. Every indexed example can be compared across source-condition, illusion, filtered-illusion, illusionless-control, and filtered-illusionless-control images.
The source-condition images are sampled from MNIST and resized to 512 × 512 pixels. Illusion images were… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/MNIST_test.IllusionAnimals_test
IllusionAnimals — Test Set
Dataset summary
This repository contains the public test split of IllusionAnimals, introduced in Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions. Each annotated example is paired across source-condition, illusion, filtered-illusion, illusionless-control, and filtered-illusionless-control conditions.
The animal source-condition images were generated with SDXL-Lightning. English scene descriptions and… See the full description on the dataset page: https://huggingface.co/datasets/VQA-Illusion/IllusionAnimals_test.DeepTumorVQA_2.0
DeepTumorVQA v2
3D abdominal-CT diagnostic Visual Question Answering benchmark with 42
clinical subtypes and 438K total QA pairs (10K curated benchmark + 428K
training pool). Includes pre-extracted 2D and video modalities, 20K agent
training trajectories with tool-use traces, and a paper-locked leaderboard.
Resources
📄 Paper (arXiv)
https://arxiv.org/abs/2605.09679
💻 Code (GitHub)
https://github.com/Schuture/DeepTumorVQA
🤗 Dataset (this… See the full description on the dataset page: https://huggingface.co/datasets/tumor-vqa/DeepTumorVQA_2.0.Japanese-Medical-VQA-12m
Japanese Medical VQA 12M
Japanese Medical VQA 12M is a large-scale Japanese medical multimodal dataset built from Open-PMC-18M and released in Parquet and Webdataset format.
This dataset contains outputs from multiple data-construction stages, including:
source captions
Japanese translations of source captions
enriched captions
Japanese translations of enriched captions
question-answering
Current Repository Format
This repository currently stores the dataset in… See the full description on the dataset page: https://huggingface.co/datasets/MIL-UT/Japanese-Medical-VQA-12m.SuperMemory-VQA
SuperMemoryVQA
SuperMemory-VQA is an egocentric visual question answering benchmark for
evaluating long-horizon memory in augmented reality assistant settings. The
dataset is designed around practical questions a person might ask a wearable
memory assistant, such as where an object was left, what someone said earlier,
whether a planned step was completed, or what happened next in a longer event.
The benchmark contains 4,853 human-verified question-answer pairs grounded in
52.9… See the full description on the dataset page: https://huggingface.co/datasets/OSU-AIoT-MLSys-Lab/SuperMemory-VQA.vqav2-smallPMC-VQA
PMC-VQA Dataset
PMC-VQA Dataset
Daraset Structure
Sample
Dataset Structure
PMC-VQA (version-1: 227k VQA pairs of 149k images).
train.csv: metafile of train set
test.csv: metafile of test set
test_clean.csv: metafile of test clean set
images.zip: images folder
(update version-2: noncompound images).
train2.csv: metafile of train set
test2.csv: metafile of test set
images2.zip: images folder
Sample
A row in train.csv is shown bellow… See the full description on the dataset page: https://huggingface.co/datasets/RadGenome/PMC-VQA.MedPix-VQA
MedPix-VQA Dataset
The MedPix-VQA dataset is a version of the data found at MEDPIX-ClinQA, specifically modified to address an image overlap issue that would result from directl splitting the original dataset. This overlap can lead to a model potentially seeing the same image during both training and validation, potentially leading to bias or data leakage.
Key Modifications:
We have modified the dataset to ensure no image overlap between the training and validation… See the full description on the dataset page: https://huggingface.co/datasets/mmoukouba/MedPix-VQA.CiQi-VQA
CiQi-Agent
Github | Model | Dataset | Paper
CiQi-Agent: Aligning Vision, Tools and Aesthetics in Multimodal Agent for Cultural Reasoning on Chinese Porcelains
Accepted to ECCV 2026
🎯 Overview
CiQi-Agent has been accepted to ECCV 2026.
We present CiQi-Agent, a domain-specific multimodal agent for antique Chinese porcelain connoisseurship. The project is designed to combine fine-grained visual perception, tool-augmented reasoning, and cultural-heritage knowledge… See the full description on the dataset page: https://huggingface.co/datasets/SII-Monument-Valley/CiQi-VQA.heico-focus-vqa
HeiCo-FOCUS (Beta release)
A clinically grounded dataset for long-context video understanding in minimally invasive surgery.
📄 Paper • 🤗 Dataset • 💻 Code • 🏆 Challenge • ⚖️ CC BY-NC-SA 4.0
[!NOTE]
Until 31 July 2026, we will employ a restricted post-release review period. During this time, we kindly ask the community to provide feedback to help us increase quality control and make continuous improvements.… See the full description on the dataset page: https://huggingface.co/datasets/orena-dkfz/heico-focus-vqa.viet-cultural-vqaVietnamese Cultural VQA Dataset is a comprehensive multimodal dataset focusing on Vietnamese cultural heritage.
It contains 28,505 images across 12 cultural categories with 119,012 question-answer pairs in Vietnamese and English.
The dataset covers diverse aspects of Vietnamese culture including architecture, cuisine, traditional clothing,
landscapes, festivals, folk culture, traditional games, sports, handicrafts, musical instruments, daily life,
and transportation.vqa
WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines
This version includes all images in the dataset. For a more lightweight and accessible alternative, please refer to the (1.1 release)[https://huggingface.co/datasets/worldcuisines/vqa-v1.1/] which reduces download size while preserving all text and metadata.
The paper was accepted to NAACL 2025 and received the Best Theme Paper award 🏆.
WorldCuisines is a… See the full description on the dataset page: https://huggingface.co/datasets/worldcuisines/vqa.VQAv2_train
Dataset Card for "VQAv2_train"
More Information needed