datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
multimodal-ct-radiology-reports
Perle AI Multi-phase CECT and CT with Radiology Reports
Summary
A de-identified CT dataset from Perle AI, paired with the original radiology reports. It supports work on multi-modal medical imaging: phase or pathology classification, report generation from images, and visual question answering.
The release has three configurations:
Config
Modality
Subjects
Pairing
cect_3phase
3-phase contrast-enhanced abdominal CT (DICOM)
5
per-subject text report +… See the full description on the dataset page: https://huggingface.co/datasets/Perle-ai/multimodal-ct-radiology-reports.svg-benchmark
Rapidata Static SVG Generation Benchmark
Built by Rapidata.
This dataset contains 1,918,367 human responses, collected with the
Rapidata Python SDK, comparing how well 42 frontier LLMs generate
static SVGs from text prompts. Each row is a head-to-head comparison between two models' renders of
the same prompt, scored by human annotators on one of three questions (Preference, Coherence, Alignment).
The SVGs are produced as raw <svg> markup by the models, rasterized to 768×768 PNGs… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/svg-benchmark.REPID
REPID: Rendering Evaluation of Photographic Image Dataset
REPID (officially introduced as the Rendering Evaluation of Photographic Image Dataset) is a large-scale benchmark designed for Image Rendering Quality Assessment (IRQA) in paper Beyond distortions: a benchmark for subjective evaluation of image rendering quality.
Unlike traditional Image Quality Assessment (IQA) which focuses on technical degradations like noise or blur, REPID aims to model subjective human aesthetic… See the full description on the dataset page: https://huggingface.co/datasets/vsevolodpl/REPID.TAIX-Ray
TAIX-Ray Dataset
TAIX-Ray is a comprehensive dataset of approximately 200k bedside chest radiographs from around 50k intensive care patients at University Hospital Aachen, Germany, collected between 2010 and 2024.
Trained radiologists provided structured reports at the time of acquisition, assessing key findings such as cardiomegaly, pulmonary congestion, pleural effusion, pulmonary opacities, and atelectasis on an ordinal scale.
Code & Details
The code for data… See the full description on the dataset page: https://huggingface.co/datasets/TLAIM/TAIX-Ray.resisc45
Description
RESISC45 dataset is a publicly available benchmark for Remote Sensing Image Scene Classification (RESISC), created by Northwestern Polytechnical University (NWPU). This dataset contains 31,500 images, covering 45 scene classes with 700 images in each class.
The dataset does not have any default splits. Train, validation, and test splits were based on these definitions here… See the full description on the dataset page: https://huggingface.co/datasets/timm/resisc45.EuroSAT_RGB
EuroSAT RGB
EUROSAT RGB is the RGB version of the EUROSAT dataset based on Sentinel-2 satellite images covering 13 spectral bands and consisting of 10 classes with 27000 labeled and geo-referenced samples.
Paper: https://arxiv.org/abs/1709.00029
Homepage: https://github.com/phelber/EuroSAT
Description
The EuroSAT dataset is a comprehensive land cover classification dataset that focuses on images taken by the ESA Sentinel-2 satellite. It contains a total of 27… See the full description on the dataset page: https://huggingface.co/datasets/blanchon/EuroSAT_RGB.eurosat-rgb
EuroSat (RGB)
Description
A dataset based on Sentinel-2 satellite images covering 13 spectral bands and consisting of 10 classes with 27000 labeled and geo-referenced samples. This is the RGB version of the dataset with visible bands encoded as JPEG images.
The dataset does not have any default splits. Train, validation, and test splits were based on these definitions here… See the full description on the dataset page: https://huggingface.co/datasets/timm/eurosat-rgb.Co-Spy-Bench
CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025)
With the rapid advancement of generative AI, it is now possible to synthesize high-quality images in a few seconds. Despite the power of these technologies, they raise significant concerns regarding misuse.
To address this, various synthetic image detectors have been proposed. However, many of them struggle to generalize across diverse generation parameters and emerging generative models.
In… See the full description on the dataset page: https://huggingface.co/datasets/ruojiruoli/Co-Spy-Bench.Defactify_Image_Dataset
Defactify_Image_Dataset
This dataset is associated with the paper A Comprehensive Dataset for Human vs. AI Generated Image Detection.
📝 Dataset Description
Dataset Summary
The Defactify_Image_Dataset (A Comprehensive Dataset for Human vs. AI Generated Image Detection) is a high-quality collection of 96,000 images and associated metadata designed to benchmark models for detecting and identifying the source of artificially generated content. Built using the MS… See the full description on the dataset page: https://huggingface.co/datasets/Rajarshi-Roy-research/Defactify_Image_Dataset.lgg-mri-segmentation-research
LGG Brain MRI Segmentation with Genomic Clusters
This repository provides a Patient-Centric version of the Lower-Grade Glioma (LGG) Segmentation dataset. While other versions of this data exist, they often treat slices as independent images. This version preserves the 3D patient volume and integrates all genomic/clinical labels directly into a multimodal-ready format.
🌟 Why This Version?
Developed for Multimodal AI Research, this dataset addresses several limitations… See the full description on the dataset page: https://huggingface.co/datasets/Ehsan-rmz/lgg-mri-segmentation-research.relaion2b-natural
LAION-Natural: Naturalness Scores for ReLAION-2B (CCN 2025, Roth & Hebart)
LAION-Natural is a large-scale naturalness scoring dataset covering 2.1 billion images from ReLAION-2B-en-research-safe. Each image receives a score predicting how "natural" or "photographic" it looks versus artificial/rendered content. At the recommended threshold of 0.7, the dataset identifies ~500 million natural photographs suitable for vision research, cognitive science, and model training.
Also… See the full description on the dataset page: https://huggingface.co/datasets/andropar/relaion2b-natural.text-2-image-Rich-Human-Feedback
Building upon Google's research Rich Human Feedback for Text-to-Image Generation we have collected over 1.5 million responses from 152'684 individual humans using Rapidata via the Python API. Collection took roughly 5 days.
If you get value from this dataset and would like to see more in the future, please consider liking it.
Overview
We asked humans to evaluate AI-generated images in style, coherence and prompt alignment. For images that contained flaws, participants were… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/text-2-image-Rich-Human-Feedback.si_us_revolutionary_era_collections
Dataset Card for Smithsonian American Revolutionary Era Collections
Dataset Summary
A specially selected subset of the Smithsonian’s Open Access collections covering objects from 1770–1810 selected for the Revolution Crossroads project in honor of the 250th anniversary of the founding of the United States. Drawn from four museums—the National Museum of American History, National Postal Museum, Smithsonian American Art Museum, and National Portrait Gallery—the… See the full description on the dataset page: https://huggingface.co/datasets/RevolutionCrossroads/si_us_revolutionary_era_collections.watercolour-reference-pool
Watercolour reference pool
The reference paintings that define the reward in the watercolour RL environment: an
agent writes a p5.brush sketch, the sketch is
rendered, and a vision judge compares the render against paintings sampled from this pool.
What the pool contains is the reward function. Replace it and you have changed what
the environment rewards, without touching a line of code.
178 paintings in two tiers, each with the JavaScript source that produced it.
tier… See the full description on the dataset page: https://huggingface.co/datasets/FineEnvs/watercolour-reference-pool.relaion2b-natural-embeddings
LAION-Natural Embeddings: CLIP ViT-H/14 Features for ~500M Natural Photographs (CCN 2025, Roth & Hebart)
LAION-Natural Embeddings provides pre-computed CLIP ViT-H/14 embeddings for ~500 million natural photographs from ReLAION-2B, filtered using the LAION-Natural naturalness classifier (score > 0.7).
Also known as: LAION-Natural Embeddings · ReLAION-Natural Embeddings · LAION-2B-Natural Embeddings
Part of the LAION-Natural dataset family, introduced in: How to sample the… See the full description on the dataset page: https://huggingface.co/datasets/andropar/relaion2b-natural-embeddings.anemia-survey-dataset
Anemia Detection — Multi-Modal Clinical SEWA Rural Dataset
Organisation: SEWA Rural — Society for Education, Welfare and Action (Rural), Jhagadia, Gujarat, India
Dataset: sewa-rural-care/anemia-survey-dataset
Contact: sewarural@ymail.com
Version: 1.0 — July 2026
Dataset Summary
This dataset supports research into non-invasive, smartphone-based anemia
screening applicable to low-resource and rural healthcare settings. It was
collected by SEWA Rural — a non-profit… See the full description on the dataset page: https://huggingface.co/datasets/sewa-rural-care/anemia-survey-dataset.watercolour-rollouts-judge-led
Watercolour rollouts, judge-led run
Browse these paintings in the gallery Space, by step and by reward, with the sketch that made each one.
Every rollout from a GRPO run that taught Qwen/Qwen3.5-35B-A3B to paint watercolours by
writing p5.brush sketches. 861 paintings, the
sketch that produced each one, and the reward it earned, indexed by training step. This
is the run with the original reward mix from the write-up, where the pairwise judge and
its hand-rated pool carry most… See the full description on the dataset page: https://huggingface.co/datasets/FineEnvs/watercolour-rollouts-judge-led.cifar100-enrichedThe CIFAR-100 dataset consists of 60000 32x32 colour images in 100 classes, with 600 images
per class. There are 500 training images and 100 testing images per class. There are 50000 training images and 10000 test images. The 100 classes are grouped into 20 superclasses.
There are two labels per image - fine label (actual class) and coarse label (superclass).rare-species
Dataset Card for Rare Species Dataset
Dataset Description
Repository: Imageomics/bioclip
Paper: BioCLIP: A Vision Foundation Model for the Tree of Life (arXiv)
Dataset Summary
This dataset was generated alongside TreeOfLife-10M; data (images and text) were pulled from Encyclopedia of Life (EOL) to generate a dataset consisting of rare species for zero-shot-classification and more refined image classification tasks. Here, we use "rare species" to mean species… See the full description on the dataset page: https://huggingface.co/datasets/imageomics/rare-species.rlbenchfail_train_dataset
Guardian: RLBench-Fail Dataset
This dataset is part of the Guardian project: Detecting Robotic Planning and Execution Errors with Vision-Language Models. It contains annotated robotic manipulation failure data generated in the RLBench simulator for training and evaluating Vision-Language Models (VLMs) on failure detection tasks.
Failures are produced by an automated pipeline that procedurally perturbs successful scripted trajectories in simulation, generating diverse planning… See the full description on the dataset page: https://huggingface.co/datasets/paulpacaud/rlbenchfail_train_dataset.watercolour-rollouts-hps-only
Watercolour rollouts, HPS-only run
Browse these paintings in the gallery Space, by step and by reward, with the sketch that made each one.
Every rollout from a GRPO run that taught Qwen/Qwen3.5-35B-A3B to paint watercolours by
writing p5.brush sketches. 470 paintings, the
sketch that produced each one, and the reward it earned, indexed by training step.
The point of the dataset is that it holds the whole run, not the good bits. Step 0 and
step 59 are both here, with the… See the full description on the dataset page: https://huggingface.co/datasets/FineEnvs/watercolour-rollouts-hps-only.retina-age-analysis
Retina Age Analysis Dataset
Dataset Description
This dataset contains 9,857 retinal fundus images from 5,393 patients for age prediction tasks.
Dataset Summary
Task: Age prediction from retinal fundus images
Images: 9,857 high-quality retinal images
Patients: 5,393 unique patients
Age Range: 5-97 years
Image Format: JPEG
Average Image Size: ~1 MB
Supported Tasks
Regression: Predict continuous age (5-97 years)
Classification: Predict age group (5… See the full description on the dataset page: https://huggingface.co/datasets/ramankamran/retina-age-analysis.700k_Human_Preference_Dataset_FLUX_SD3_MJ_DALLE3
NOTE: A newer version of this dataset is available Imagen3_Flux1.1_Flux1_SD3_MJ_Dalle_Human_Preference_Dataset
Rapidata Image Generation Preference Dataset
This Dataset is a 1/3 of a 2M+ human annotation dataset that was split into three modalities: Preference, Coherence, Text-to-Image Alignment.
Link to the Coherence dataset: https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Coherence_Dataset
Link to the Text-2-Image Alignment dataset:… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/700k_Human_Preference_Dataset_FLUX_SD3_MJ_DALLE3.Runway_Frames_t2i_human_preferences
Rapidata Frames Preference
This T2I dataset contains roughly 400k human responses from over 82k individual annotators, collected in just ~2 Days using the Rapidata Python API, accessible to anyone and ideal for large scale evaluation.
Evaluating Frames across three categories: preference, coherence, and alignment.
Explore our latest model rankings on our website.
If you get value from this dataset and would like to see more in the future, please consider liking it.… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/Runway_Frames_t2i_human_preferences.mini-reachy-animation
Reachy Mini Animation Dataset
Multi-view renders of 85 emotional animations performed by the Reachy Mini
robot, paired with the full robot joint state for every single frame.
Source of the animations. The emotional animations rendered here come from the
official pollen-robotics/reachy-mini-emotions-library
dataset by Pollen Robotics. This dataset re-renders those emotions from 12 camera
angles (with 3 background variants) and pairs every frame with the robot's joint state.… See the full description on the dataset page: https://huggingface.co/datasets/BastienATOS/mini-reachy-animation.bakkhali-river-high-low-tide
Bakkhali River — High Tide vs Low Tide, Bangladesh
517 photographs of the Bakkhali River near Cox's Bazar, Bangladesh, documenting the same general stretch of river at high tide (264 images) and low tide (253 images). Captured across 10 separate sessions between 2 July and 15 August 2026.
This is not a frame-by-frame matched pair set — sessions were shot on different dates and the camera position varies within each session — but high- and low-tide frames come from the same short… See the full description on the dataset page: https://huggingface.co/datasets/golamrob/bakkhali-river-high-low-tide.human-style-preferences-images
Rapidata Image Generation Preference Dataset
This dataset was collected in ~4 Days using the Rapidata Python API, accessible to anyone and ideal for large scale data annotation.
Explore our latest model rankings on our website.
If you get value from this dataset and would like to see more in the future, please consider liking it.
Overview
One of the largest human preference datasets for text-to-image models, this release contains over 1,200,000 human preference… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/human-style-preferences-images.OpenAI-4o_t2i_human_preference
Rapidata OpenAI 4o Preference
This T2I dataset contains over 200'000 human responses from over ~45,000 individual annotators, collected in less than half a day using the Rapidata Python API, accessible to anyone and ideal for large scale evaluation.
Evaluating OpenAI 4o (version from 26.3.2025) across three categories: preference, coherence, and alignment.
Explore our latest model rankings on our website.
If you get value from this dataset and would like to see more in the… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/OpenAI-4o_t2i_human_preference.Flux_SD3_MJ_Dalle_Human_Alignment_Dataset
NOTE: A newer version of this dataset is available Imagen3_Flux1.1_Flux1_SD3_MJ_Dalle_Human_Alignment_Dataset
Rapidata Image Generation Alignment Dataset
This Dataset is a 1/3 of a 2M+ human annotation dataset that was split into three modalities: Preference, Coherence, Text-to-Image Alignment.
Link to the Coherence dataset: https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Coherence_Dataset
Link to the Preference dataset:… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Alignment_Dataset.AEGISThis repository contains the data of the paper [AEGIS: A Holistic Benchmark for Evaluating Forensic Analysis of AI-Generated Academic Images]