Loading...
Loading...
Image-text pairs, instruction tuning, visual QA, cross-modal data, foundation model training data
1,932 datasets
Ayn-VQA is a multimodal evaluation dataset for culturally specific Arabic image understanding. It was created by QCRI as part of the ImageEval 2026 Shared Task at ArabicNLP 2026 and was last updated on June 4, -2026. The dataset tests a model's ability to answer questions about images and distinguish grounded descriptions from plausible hallucinations, with tasks offered in both English and Modern Standard Arabic.
Compar:IA is a public chatbot arena run by the French Ministry of Culture. This dataset contains French-language conversations where users compare two anonymous models' answers and state their preference. Each row likely contains one turn of a conversation, including the two model responses and the user's indicated preference.
A multimodal dataset of natural conversations between pairs of friends, captured with two simultaneous camera angles and a separate audio recording. Each sample is a side-by-side video combining both camera views with the original audio track. The dataset was created by liva-ai and last updated on 2026-06-09.
GUIDE (GUI User Intent Detection Evaluation) is a benchmark for evaluating multimodal models on perceiving user behavior and inferring intent in open-ended GUI tasks. It consists of 67.5 hours of screen recordings from 120 novice user demonstrations with think-aloud narrations, across 10 software applications. The dataset was created by Saelyne Yang, Jaesang Yu, Yi-Hao Peng, Kevin, and others, and was last updated on Hugging Face in June 2026.
MMTutorBench is the first multimodal benchmark for AI math tutoring, containing 770 carefully curated problems paired with 1,414 images. The dataset provides structured reference answers and per-instance rubrics for evaluating large language models along three pedagogical axes: Insight, Operation Formulation, and Operation Execution. It was created by Tangchiu and last updated on May 22, 2026.
VLM Eval Videos is a benchmark dataset containing 693 short MP4 video clips for evaluating Vision-Language Models. The dataset, created by author gnitoahc, is organized into five categories, with each clip paired with a fixed question and a ground-truth short-sentence answer. It was last updated on the Hugging Face platform in June 2026.
Over 15 million data points covering math, code, knowledge, and instruction following form the full set of core-domain SFT data used for post-training the MiniCPM5-1B-SFT model. This dataset is a key representative of L3 refined data within the UltraData L0-L4 tiered data management framework. It was authored by openbmb and last updated on Hugging Face in May 2026.
A research protocol for a cross-sectional study comparing 30 individuals with schizophrenia and 30 healthy controls. The study, authored by Peipei Luan, involves behavioral interoceptive tasks, clinical symptom ratings, cognitive testing, and multimodal MRI scanning. The protocol was last updated on 2026-04-17.
A multimodal dataset combines sequential CTA slices and six clinical biomarkers from 263 symptomatic abdominal aortic aneurysm patients. It was created by Jiaxin Cheng for developing and validating a deep learning model for rupture risk assessment. The dataset was last updated in April 2026.
A retrospective cohort of 263 symptomatic abdominal aortic aneurysm (AAA) patients, with data split into a 230-patient development cohort and a 33-patient independent temporal test set. This multimodal dataset combines sequential computed tomography angiography (CTA) slices with six key clinical biomarkers, created by Jiaxin Cheng and published in April 2026 to develop a deep learning model for predicting impending rupture.
Yanziang's InternVideo3 dataset provides 380,000 samples for supervised fine-tuning of models on long-video understanding. Each sample contains a YouTube video ID paired with question-answer annotations for detailed description and reasoning tasks. The dataset was uploaded to Hugging Face and last updated on June 11, 2026.
ObsCrisis-Bench is a multimodal benchmark containing 4,599 visual question-answering samples for evaluating large vision-language models. The dataset covers 127 extreme weather events across 8 disaster categories, combining satellite multispectral imagery with optional weather station data. It was created by author YYQ898 and last updated on the Hugging Face platform in June 2026.
KoSum is a benchmark dataset containing 700 recent Korean YouTube videos uploaded between 2024 and 2026, spanning 14 fine-grained content categories. It provides per-second visual, audio, and text features aligned with YouTube Most-Replayed importance scores. The dataset was created by author iontail and last updated on June 14, 2026.
UniReason-Med Data is a training dataset for the UniReason-Med medical multimodal reasoning framework. The associated paper describes a framework for 2D and 3D medical image understanding with interleaved image-text reasoning and region grounding. The dataset was released by IQuestLab and was last updated on the Hugging Face platform in June 2026.
TRM-Preference is a dataset introduced in the paper 'Characterizing, Evaluating, and Optimizing Complex Reasoning' for evaluating and optimizing reasoning traces in Large Reasoning Models. The dataset applies the ME² principle to assess 'how a model thinks' across dimensions like Macro-Efficiency. It was authored by zzzhr97 and last updated on Hugging Face in June 2026.
100 tabular samples of synthetically generated Electroencephalography (EEG) features, designed for testing and optimizing closed-loop neuro-generative software architectures. The data provides pre-extracted Power Spectral Density (PSD) features ready for direct ingestion by machine learning classifiers. It was authored by Shubham Sunil Kumar and last updated on June 29, 2026.
A set of benchmark functions used to evaluate the Felis Catus Optimization (FCO) metaheuristic algorithm. The dataset includes functions from the CEC 2005 and CEC 2017 benchmark suites. It was created by Mohammad Salehi and last updated on 2026-04-15.
A set of benchmark functions used to evaluate the Felis Catus Optimization (FCO) metaheuristic algorithm. The dataset includes results from experiments on the CEC 2005 and CEC 2017 benchmark suites and three real-world engineering design problems. It was created by Mohammad Salehi and last updated on April 15, 2026.
JumpShift is a Persian-language alignment dataset developed by JumpLander for training coding assistants and software engineering agents. It is designed to help models respond like professional coding assistants, focusing on reasoning, clarification, and safe behavior. The dataset was last updated on June 13, 2026.
Counterfactual VLM Benchmark Data is a dataset payload for evaluating vision-language models. It was authored by JingyuSun and uploaded to Hugging Face on May 20, 2026. The dataset is intended to be used with an associated GitHub repository containing download and evaluation scripts.