Loading...
Loading...
Image-text pairs, instruction tuning, visual QA, cross-modal data, foundation model training data
1,929 datasets
RoboShackles is a safety benchmark for evaluating Embodied Foundation Models. The public test split contains 1,200 safety-critical robotic video clips, with 200 videos per category. The dataset was created by YZW00 and last updated on Hugging Face in June 2026.
OpenBrush Landscapes is a curated subset of the OpenBrush-75K dataset containing every landscape painting from the parent collection. It includes 12,612 images across all artists, movements, and centuries, curated so users do not need to download the full 75,313-image dataset. The subset was created by jaddai and was last updated on May 27, 2026.
6,119 religious paintings curated from the OpenBrush-75K collection. The dataset focuses on saints, biblical scenes, and devotional works, with a heavy emphasis on Renaissance and Baroque eras. It was created by jaddai and last updated on May 27, 2026.
4.3 GB of thermal recordings from controlled laboratory experiments on high-impedance faults in medium-voltage covered conductors. Diogo Biasuz Dahlke created the dataset, which includes thermal video files (MP4) and native radiometric files (HRV) capturing temperature evolution during fault initiation. The dataset was last updated on 2026-05-05.
13,059 portrait paintings curated from the OpenBrush-75K dataset, spanning artistic movements from Renaissance to Realist. The subset was created by jaddai using the Qwen3-VL-30B-A3B vision-language model and last updated on May 27, 2026. It provides a focused collection of portraits under a CC0 license.
IndustryBench-MIPU is a benchmark dataset for evaluating multimodal large language models on extracting product specifications from multiple heterogeneous images. The dataset, created by alibaba-multimodal-industrial-ai, tests model capabilities in text recognition, visual reasoning, domain knowledge, and cross-image evidence integration. It was last updated on June 15, 2026.
A research dataset containing performance metrics for a multimodal conversational agent named EAC-Agent. The dataset likely contains results from validation on benchmark datasets IEMOCAP and MELD. It was uploaded by Shahid Jamil to figshare on 2026-04-17.
A multimodal dataset for visual question answering in additive manufacturing, focusing on quality assessment of Ti10Mo6Cu alloy parts produced via Laser Powder Bed Fusion. The dataset was created by AI4Manufacturing and was last updated on July 16, 2026. Each row contains fields for a query, an image, an annotation, reasoning, category, task, and metadata.
4,240 Baroque-era artworks curated from the larger OpenBrush-75K collection. The subset focuses on the canonical Baroque visual language from approximately 1600 to 1750, characterized by chiaroscuro and dramatic lighting. It was created by jaddai and last updated on May 27, 2026.
UniCure is a multi-modal framework integrating omics and chemical foundation models to predict transcriptomic drug responses. This repository contains the pre-processed datasets, configuration files, and pre-trained model weights required to reproduce the results. The archive is 12.4 GB and was last updated on 2026-04-23 by Zexi Chen.
A benchmark dataset for evaluating whether a Large Language Model correctly allocates verification when orchestrating a fallible biology foundation model. The dataset, created by jang1563, includes a substrate table from the GEARS/Norman experiment. It was last updated on June 17, 2026.
CRYSTAL is a diagnostic benchmark for evaluating multimodal reasoning step by step, not just by the final answer. Each instance pairs an image and a question with an ordered sequence of natural-language reference reasoning steps, enabling step-level metrics like Match F1 and Ordered Match F1 alongside answer accuracy. The dataset was created by author waybarrios and was last updated on the platform in June 2026.
1,889 images of Vincent van Gogh's works, curated from the larger OpenBrush-75K collection. All images are paired with structured captions generated by the Qwen3-VL-30B-A3B vision-language model. The dataset was created by jaddai and last updated on May 27, 2026.
221 trials from 11 novice participants manipulating objects with 20 visually encoded fragility levels (50–1000 gf). The dataset includes synchronized multimodal observations: robot joint trajectories, gripper force signals, multi-view RGB video, users' perceived fragility, confidence ratings, and trial outcomes. It was released by Jin Ong on figshare in April 2026.
LeafBench is a visual question answering benchmark derived from the LeafNet dataset. It is designed to evaluate Vision-Language Models on six hierarchical diagnostic tasks for plant diseases. The dataset was created by author 'enalis' and was last updated on June 20, 2026.
A curated subset of 1,334 Claude Monet artworks from the OpenBrush-75K collection. The images are paired with structured captions generated by the Qwen3-VL-30B-A3B vision-language model. This dataset was created by jaddai and last updated on May 27, 2026.
188 septic patient records, including 89 with sepsis-associated acute kidney injury, integrate clinical variables with transcriptomic-guided blood biomarkers. Weiqin Wu developed this dataset to build a multimodal predictive model, with data last updated in April 2026. It features predictors like SOFA score, MAP, BUN, CRP, and the immune biomarkers CD177 and IL18R1.
A 2026 study by Lena Rommerskirchen applied machine learning to multimodal data from 110 participants with bulimia nervosa, binge eating disorder, and matched controls. The dataset integrates task-based fMRI, intrinsic connectivity, voxel-based morphometry, neuropsychological assessments, and peripheral blood biomarkers. It was used to classify diagnostic groups and predict individual symptom variation.
Data Sheet 3 presents multimodal data from a study of 110 participants with bulimia nervosa, binge eating disorder, and matched controls. The dataset integrates task-based fMRI, intrinsic connectivity, voxel-based morphometry, neuropsychological assessments, and peripheral blood biomarkers. It was authored by Lena Rommerskirchen and last updated on figshare in April 2026.
MNIST-VQA is a synthetic Visual Question Answering dataset generated from MNIST digits placed on a 3x3 grid. It is designed to test spatial reasoning, object localization, counting, and existence verification capabilities of VQA models. The dataset was created by author star092304 and was last updated on the Hugging Face platform in June 2026.