datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
RealworldQA
RealWorldQA
RealWorldQA is a benchmark designed for real-world understanding. The dataset consists of anonymized images taken from vehicles, in addition to other real-world images. We are excited to release RealWorldQA to the community, and we intend to expand it as our multimodal models improve.
The initial release of the RealWorldQA consists of over 700 images, with a question and easily verifiable answer for each image. See the announcement of Grok-1.5 Vision Preview.… See the full description on the dataset page: https://huggingface.co/datasets/xai-org/RealworldQA.vlmsareblindArXiv - Website
X-AIGD
X-AIGD
X-AIGD is a fine-grained benchmark designed for eXplainable AI-Generated image Detection. It provides pixel-level human annotations of perceptual artifacts in AI-generated images, spanning low-level distortions, high-level semantics, and cognitive-level counterfactuals, aiming to advance robust and explainable AI-generated image detection methods.
For more details, please refer to our paper: Unveiling Perceptual Artifacts: A Fine-Grained Benchmark for Interpretable… See the full description on the dataset page: https://huggingface.co/datasets/Coxy7/X-AIGD.Cabin-Human-Behavior-Dataset
全球最大的智能座舱多模态开源高质量数据集来啦！
一. 数据集摘要 (Dataset Summary)
「CyberData塞塔」智能座舱用户行为数据集是一个专为加速智能座舱感知算法开发而设计的高质量、程序化生成的图像数据集。随着 C-NCAP、EU GSR 等全球汽车安全法规对驾驶员监控系统 (DMS) 和乘客监控系统 (OMS) 提出更高要求，安全、合规、多样化的训练数据变得至关重要。本数据集通过合成方式，旨在解决真实世界数据采集面临的隐私风险、高昂成本和长尾场景覆盖不足等核心挑战。
该数据集包含 5,000 张 由 XAI Lab 自主研发的数据集生成引擎合成的高保真座舱内用户行为图像，每张图像都附带丰富的、100% 精确的标注信息。
核心特点:
丰富的场景多样性: 涵盖不同年龄、性别、种族和衣着风格的虚拟人模型，以及多种驾驶与乘坐行为（如使用手机、喝水、疲劳、手势）和面部表情。
专为座舱感知优化: 数据集可直接用于智能座舱端侧视觉模型，尤其是 DMS/OMS 算法的训练、微调与验证，帮助模型精准理解座舱内复杂的交互与状态。… See the full description on the dataset page: https://huggingface.co/datasets/XAILab-CyberSpark/Cabin-Human-Behavior-Dataset.xAI_Aurora_t2i_human_preferences
Rapidata Aurora Preference
This T2I dataset contains over 400k human responses from over 86k individual annotators, collected in just ~2 Days using the Rapidata Python API, accessible to anyone and ideal for large scale evaluation.
Evaluating Aurora 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/xAI_Aurora_t2i_human_preferences.Cabin-Human-ABNORMAL-Behavior-Dataset
全球最大的智能座舱多模态开源高质量数据集来啦！
一. 数据集摘要 (Dataset Summary)
「CyberData塞塔」智能座舱用户行为数据集是一个专为加速智能座舱感知算法开发而设计的高质量、程序化生成的图像数据集。随着 C-NCAP、EU GSR 等全球汽车安全法规对驾驶员监控系统 (DMS) 和乘客监控系统 (OMS) 提出更高要求，安全、合规、多样化的训练数据变得至关重要。本数据集通过合成方式，旨在解决真实世界数据采集面临的隐私风险、高昂成本和长尾场景覆盖不足等核心挑战。
该数据集包含 5,000 张 由 XAI Lab 自主研发的数据集生成引擎合成的高保真座舱内用户行为图像，每张图像都附带丰富的、100% 精确的标注信息。
数据格式
数据集以JSON格式提供，包含以下字段：
image_id: 图像ID
image_path: 图像路径
category: 行为类别
tags: 行为标签
behaviors: 包含左右乘客行为描述的对象
left_passenger: 左侧乘客行为描述… See the full description on the dataset page: https://huggingface.co/datasets/XAILab-CyberSpark/Cabin-Human-ABNORMAL-Behavior-Dataset.OpenImages-Inpaintedxai-attack-detection-cifar10
XAI Attack Detection — CIFAR-10 PGD
This private research dataset contains balanced, paired clean and adversarial images for
studying whether an attack can be detected from a classifier explanation map.
Dataset construction
The source is the CIFAR-10 test split. A fine-tuned OpenCLIP ViT-B/16 classifies each
clean image. Clean-correct examples are attacked with untargeted L-infinity PGD using
epsilon 8/255, step size 2/255, 10 steps, and deterministic random… See the full description on the dataset page: https://huggingface.co/datasets/nimaeb/xai-attack-detection-cifar10.X-AIGD-demoThis is a tiny demo subset of X-AIGD for debugging purposes only. Please visit the dataset card of X-AIGD for descriptions.
XAIGID-RewardBench
XAIGID-RewardBench
The test split contains 3,988 benchmark triplets with an image, two policy-model responses, and up to three human-response slots. It contains 4,412 populated human annotations. The human_response split contains the 336-image Human-vs-Policy subset, including 36 GPT-5.5 comparisons, in the same image-and-response schema.
See the project repository and paper.
Xaixai-attack-detection-imagenette
XAI Attack Detection: Imagenette targeted BIM/PGD on ViT-B/16
Private research dataset of paired clean and targeted adversarial Imagenette images. It is
built to study how adversarial attacks change a Vision Transformer's explanation maps and to
support later work on attack detection. Each row is one source image with its clean and its
attacked version.
Summary
Pairs
12,420 (train 8,690 · validation 1,860 · test 1,870)
Source images
Imagenette v2… See the full description on the dataset page: https://huggingface.co/datasets/nimaeb/xai-attack-detection-imagenette.layoutflow-xai
Query-Driven Attribution Framework for Iterative Layout Generation (LayoutFlow XAI)
These are the results produced by the official implementation of the 2026 paper “Query-Driven Attribution Framework for Iterative Layout Generation”, where we apply the framework to the LayoutFlow method and use Integrated Gradients as the attribution method.
To view the results in a more organized way, switch to the “Files and versions” tab.
Code Repository
For the official code… See the full description on the dataset page: https://huggingface.co/datasets/vaarga/layoutflow-xai.PCNN-StanfordDogs-HumanStudy-Datanew_conv_xai_augmented
Dataset Description
This dataset is derived from new_conv_xai. new_conv_xai had the integration of all the images and other metadata together with each of the 30 conversations from the train and test json files.
new_conv_xai_augmented takes it a step further and converts each of those conversations into subsets of conversation histories, creating inputs and outputs that we can finetune the model on.
Each of these input-output pairs are associated with their corresponding… See the full description on the dataset page: https://huggingface.co/datasets/mxforml/new_conv_xai_augmented.Cabin-multi-modal-dataset-of-sightseeing
由XAI Lab 数据集生成引擎合成，主要面向汽车智能座舱端侧模型微调训练，内容为汽车舱外景色图像
It is synthesized by the XAI Lab data set generation engine, mainly for the fine-tuning training of the end-side model of the vehicle intelligent cockpit, and the content is the image of the scene outside the vehicle
license: apache-2.0
language:
- zh
tags:
- Cabin
- Sightseeing
- Camera
- AIGC
twitternew_conv_xaiX-AIGD-real
X-AIGD (real images)
Dataset Structure
The X-AIGD dataset is split into two components:
Generated Images: Available publicly in the main X-AIGD repo.
Real Images: The metadata and download script are provided in the main repo. A subset of image copies is hosted in this gated repository for research reproducibility.
Real Image Sources
COCO: Provided here as an archive for research reproducibility. Note that the image copies here differ from the original… See the full description on the dataset page: https://huggingface.co/datasets/Coxy7/X-AIGD-real.visliesCHM-Corr-Datacleaned_conv_xai_augmented
Dataset Description
This dataset is a recreation of new_conv_xai, after I went back and completely cleaned up the original train and test jsons.
This dataset is derived from cleaned_conv_xai. cleaned_conv_xai had the integration of all the images and other metadata together with each of the 30 conversations from the train and test json files.
cleaned_conv_xai_augmented takes it a step further and converts each of those conversations into subsets of conversation histories, creating… See the full description on the dataset page: https://huggingface.co/datasets/mxforml/cleaned_conv_xai_augmented.cleaned_conv_xaiXAI_CropDiseaseDetection