datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Spatial-DISE
Spatial-DISE: A Unified Benchmark for Evaluating Spatial Reasoning in Vision-Language Models
📋 Overview
Spatial-DISE is a comprehensive benchmark dataset designed to evaluate spatial reasoning capabilities in vision-language models. The dataset focuses on various aspects of spatial intelligence including 3D perception, spatial transformation, and geometric reasoning across multiple difficulty levels.
🧪 Evaluation Support
Supported:… See the full description on the dataset page: https://huggingface.co/datasets/TACPS-liv/Spatial-DISE.llbench-dataset
LL-Bench: Rethinking Low-Level Vision Evaluation in the Era of Large-Scale Generative Models via Human Preferences
Anonymous release prepared for NeurIPS 2026 review. Please do not redistribute.
LL-Bench is a large-scale, human-preference benchmark for evaluating low-level
vision restoration in the era of large generative models (LGMs). It compares
10 LGMs with 16 specilist and 5 all-in-one models across 16 low-level vision tasks, paired with dense human annotations:pairwise… See the full description on the dataset page: https://huggingface.co/datasets/anonymousllbench/llbench-dataset.localizer
Dartbrains Localizer Dataset
A subset of the Brainomics/Localizer functional MRI dataset, prepared for the Dartbrains neuroimaging course at Dartmouth College.
Quick Start
Load beta maps (recommended for most exercises)
from datasets import load_dataset
ds = load_dataset("dartbrains/localizer", "betas")
img = ds[0]["nifti"] # nibabel.Nifti1Image
subject = ds[0]["subject"] # "S01"
condition = ds[0]["condition"] # "audio_computation"… See the full description on the dataset page: https://huggingface.co/datasets/dartbrains/localizer.multimodal-LLMs-See-Sentiment
MLLMsent — datasets and experiment results
Every input and every output of "Multimodal LLMs See Sentiment"
(arXiv:2508.16873): the image descriptions generated by six multimodal
LLMs, the sentiment labels derived from the PerceptSent annotations, and the complete
per-fold results of all 141 experiments.
Paper: arXiv:2508.16873
Code, training and inference: https://github.com/neemiasbsilva/multimodal-LLMs-see-sentiment
Model checkpoints:… See the full description on the dataset page: https://huggingface.co/datasets/neemiasbsilva/multimodal-LLMs-See-Sentiment.ridgelora-cross-sensor-sd302d-f-to-m-20260825
RidgeLoRA-FP: SD302A-F to SD302D-M cross-sensor experiment
This public archive contains the leakage-controlled direct cross-sensor
experiment used to evaluate whether Stage-2 synthetic target-sensor images
help recognition on a physically different real sensor.
Locked protocol
Source/condition sensor: NIST SD302A device F.
Target sensor: NIST SD302D device M.
Identity: subject:finger-position; the same fingers exist across both
collections.
Subject split: 160… See the full description on the dataset page: https://huggingface.co/datasets/LamTNguyen/ridgelora-cross-sensor-sd302d-f-to-m-20260825.bigearthnet
BigEarthNet - HDF5 version
This repository contains an export of the existing BigEarthNet dataset in HDF5 format. All Sentinel-2 acquisitions are exported according to TorchGeo's dataset (120x120 pixels resolution).
Sentinel-1 is not contained in this repository for the moment.
CSV files contain for each satellite acquisition the corresponding HDF5 file and the index.
A PyTorch dataset class which can be used to iterate over this dataset can be found here, as well as the script used… See the full description on the dataset page: https://huggingface.co/datasets/lc-col/bigearthnet.honeybee-samples
HoneyBee Sample Files
Sample data and resource files for the HoneyBee framework — a scalable, modular toolkit for multimodal AI in oncology.
These files are used by the HoneyBee example notebooks (clinical, pathology, radiology) and by HoneyBee's molecular processing code at runtime (Hugo_symbols.tsv is fetched on first use of DNA mutation preprocessing).
Paper: HoneyBee: A Scalable Modular Framework for Creating Multimodal Oncology Datasets with Foundational Embedding Models… See the full description on the dataset page: https://huggingface.co/datasets/Lab-Rasool/honeybee-samples.WorcesterMA_Housing_Facades
WorcesterMA_Housing_Facades:
🌐 GitHub | 🤗 Dataset
Street-level photographs of housing facades from Worcester, MA, organized into four facade classes. Each image filename is the property PID (integer). The dataset links housing registry metadata (e.g., year_built) with facade images collected for research in visual housing classification.
Dataset Card
Dataset name: WorcesterMA_Housing_Facades
Short description: Photographs of housing facades from Worcester, MA.… See the full description on the dataset page: https://huggingface.co/datasets/murai-lab/WorcesterMA_Housing_Facades.Emotion_Video_Facial_Landmarks
Dataset Card for 478-Point Normalized 3D Facial Landmark Dataset
Dataset Description
This dataset provides pre-extracted, normalized 3D facial landmark features derived from the Video Emotion dataset. It is optimized for efficient training of emotion recognition and facial analysis models, bypassing the need to process large raw video files.
License: The extracted feature data in this CSV file is licensed under Apache 2.0. Note that the original source video files may… See the full description on the dataset page: https://huggingface.co/datasets/PSewmuthu/Emotion_Video_Facial_Landmarks.ad-creative-quality-human-vs-llm
Human Expert vs LLM Judge: Facebook Ad Creative Quality
500 real Facebook ads from 253 advertisers, each rated for creative quality by a human ad expert AND by a vision LLM — with the LLM's full reasoning.
The headline finding baked into this data: the human and the LLM agree on image quality only 26.8% of the time. The LLM judge rates 71.8% of ads "good"; the human expert rates only 20% "good". If you are using an LLM as a judge of ad creative (or any subjective visual quality)… See the full description on the dataset page: https://huggingface.co/datasets/AdControlCenter/ad-creative-quality-human-vs-llm.Latent-Resonance-AI-Image-Forensics-Benchmark-N1000
Latent Resonance: SOTA Large-Scale AI Image Forensics Benchmark (N=1,000)
Author: Debdip Bandyopadhyay (Independent AI Researcher, Kolkata, India; M.Tech, IIT Jodhpur, AI & Data Science)Preprint & Paper: Latent Resonance: Zero-Shot Autoencoder Inversion and Azimuthal Spectral Forensics for Diffusion Image Attribution (IEEE Flagship / CERN Zenodo 2026)
1. Executive Summary & Diagnostic Suite
This repository contains the complete empirical evaluation records… See the full description on the dataset page: https://huggingface.co/datasets/DebdipCS/Latent-Resonance-AI-Image-Forensics-Benchmark-N1000.osm-europe-1k
OSM-Europe-1k
A 1,000-image street-level geolocation benchmark for Europe, sampled from
the OpenStreetView-5M (OSV-5M)
test split. Intended as a contamination-free, openly-licensed reference set
for evaluating image→GPS models. Companion benchmark to a master's thesis at
FH JOANNEUM (Florian Leber, 2026).
What's in this repo:
The 1,000 OSM/Mapillary image bytes are bundled directly under
images/ (66 MB) — re-distribution is allowed by the upstream
CC-BY-SA 4.0 license, with… See the full description on the dataset page: https://huggingface.co/datasets/lebfla11/osm-europe-1k.europe-holdout-2k
Europe-Holdout-2k
A 2,000-location street-level geolocation benchmark sampled from the
held-out test fold of the thesis training pipeline. Designed as a
high-variation, contamination-free probe for the published thesis models
(v14 unfreeze_last2, v15, ...) and any external geolocation system.
Locations
2,000
Region
Europe (42 countries)
Source
seed=42 location-level 80/10/10 split of the thesis training corpus
Imagery
Coordinates only — Google Street View bytes… See the full description on the dataset page: https://huggingface.co/datasets/lebfla11/europe-holdout-2k.IMAGE_SCENE_LEBELLINGEmotion_Video_Facial_Landmarks
Dataset Card for 478-Point Normalized 3D Facial Landmark Dataset
Dataset Description
This dataset provides pre-extracted, normalized 3D facial landmark features derived from the Video Emotion dataset. It is optimized for efficient training of emotion recognition and facial analysis models, bypassing the need to process large raw video files.
License: The extracted feature data in this CSV file is licensed under Apache 2.0. Note that the original source video files may… See the full description on the dataset page: https://huggingface.co/datasets/mac26/Emotion_Video_Facial_Landmarks.beyond_the_lab_neurips_paperhumanoid-pose-state-dataset-lite
Humanoid Pose State Dataset Lite
Lightweight synthetic dataset for humanoid robot pose classification.
Pose Classes
neutral
walking_pose
running_pose
sitting_pose
lifting_pose
waving_pose
Structure
dataset/
├── train/
├── validation/
Each split contains pose-labeled image folders.
Total Samples
Train: 600
Validation: 150
Image Format
RGB, 224x224
License
MIT
ArabicImageNet
Arabic Translated ImageNet Dataset
Overview
Welcome to the Arabic Translated ImageNet Dataset! This dataset is a version of the ImageNet dataset, translated into Arabic. ImageNet is a widely used benchmark for image classification tasks, and this translation aims to facilitate research and development in the Arabic language.
Contents
Arabic_query_short: Arabic translations of the original ImageNet class labels.
Arabic_query_long: Arabic translations of the… See the full description on the dataset page: https://huggingface.co/datasets/LinaAlhuri/ArabicImageNet.landmark-swin-pretrained