datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
vlm_evaluation_v1.0
Datacard
This dataset is the evaluation VLM dataset used in VLABench. It is designed to evaluate the planning capabilities of Vision-Language Models (VLMs) in embodied scenarios.
Source
Project Page: https://vlabench.github.io/
Arxiv Paper: https://arxiv.org/abs/2412.18194
Code: https://github.com/OpenMOSS/VLABench
Uses
The dataset structure is as follows:
vlm_evaluation_v1.0/
├── CommenSence/
├── add_condiment_common_sense/
├──… See the full description on the dataset page: https://huggingface.co/datasets/VLABench/vlm_evaluation_v1.0.Chat2Workflow-Evaluation
Chat2Workflow
Chat2Workflow is a benchmark designed for evaluating the ability of Large Language Models (LLMs) to generate executable visual workflows from natural language instructions.
Paper: Chat2Workflow: A Benchmark for Generating Executable Visual Workflows with Natural Language
Repository: zjunlp/Chat2Workflow
Overview
Executable visual workflows are widely used in industrial deployments for their reliability and controllability. Chat2Workflow addresses the… See the full description on the dataset page: https://huggingface.co/datasets/zjunlp/Chat2Workflow-Evaluation.Egocentric_10K_Evaluation
Dataset Card for Egocentric_10K_Evaluation
This is a FiftyOne dataset with 30000 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/Egocentric_10K_Evaluation")
# Launch the App
session = fo.launch_app(dataset)
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/Egocentric_10K_Evaluation.recap-t2i-evaluation-sample-2026
Recaptioned T2I Supervision Evaluation Sample
This repository is the small reviewer-inspection companion to the full anonymous caption-metadata release. The full release is hosted separately at https://huggingface.co/datasets/Anonymous1477/recap-t2i-evaluation-metadata-2026; this repository stays under the large-dataset sample threshold and gives reviewers a direct way to inspect redacted caption metadata, join structure, and selected image-conditioned audit packages.… See the full description on the dataset page: https://huggingface.co/datasets/Anonymous1477/recap-t2i-evaluation-sample-2026.DL3DV-Evaluation
DL3DV Testing Split Download Instructions
This repo contains all 55 scenes for evaluation. Note: it is an independent dataset, and none of its scenes overlap with those in DL3DV-10K. Have a galance on the preview page: https://dl3dv-10k.github.io/DL3DV-Testing-Split-Preview/.
Download
As the whole benchmark dataset is ~500G, a python script to download and untar files.
Environment Setup
The download script relies on huggingface hub, tqdm. You can download by… See the full description on the dataset page: https://huggingface.co/datasets/DL3DV/DL3DV-Evaluation.artifacts_evaluation
GRADE artifact evaluation (MobiCom 2026)
This anonymous artifact reproduces the quantitative evaluation of the GRADE paper. The release provides 19 model checkpoints plus one auxiliary TAESD weight file (20 .safetensors files in total), code for local inference and metric computation, and the saved camera-ready quantitative results.
Two evaluation paths are supported:
E1 — saved-result reproduction: regenerate the quantitative tables and figures from the supplied camera-ready… See the full description on the dataset page: https://huggingface.co/datasets/mypersonalsharingspot11/artifacts_evaluation.NuosuBburma-OCR-Evaluation-Set
NuosuBburma OCR Evaluation Set
规范彝文 OCR 评估集
用于规范彝文（NuosuBburma）的模型性能评估与错误分析。
内容涵盖纯彝文、彝汉混排及少量含拉丁字母的混排文本；场景覆盖书籍、手写、屏幕和实拍。
真实性声明：全部评估样本来自真实扫描或实际拍摄，不含合成评估数据。
评估任务
输入为单张包含规范彝文、彝汉混排及少量拉丁字母、数字或标点的图像。模型按照图像中的视觉阅读顺序转写可见文字，保留必要的行结构和原有字符，不进行翻译、改写或文本补全。评估重点是复杂版式、混排文字和真实拍摄干扰条件下的文字识别能力。
在线可视化
查看评估集分布与图像—标准答案对照
内容统计
项目
数量
样本 / 图片
1030 / 1030
文档页 / OCR 实例
519 / 511
图像类别
类别
数据形态
主要干扰
评估重点… See the full description on the dataset page: https://huggingface.co/datasets/nanxidajun/NuosuBburma-OCR-Evaluation-Set.Egocentric-10K-Evaluation
Egocentric10K
Ego4D
Epic-Kitchens
To evaluate the three in-the-wild egocentric datasets Egocentric-10K, Ego4D, and EPIC-KITCHENS-100 on hand visibility and active manipulation density as a proxy for data efficiency, we randomly sample 10k frames from each dataset and run them through a gemini-2.5-flash.
Hand Visibility
Prompt:
You are labeling an egocentric first-person image. Your task is to count… See the full description on the dataset page: https://huggingface.co/datasets/builddotai/Egocentric-10K-Evaluation.monocular-geometry-evaluationProcessed versions of some open-source datasets for evaluation of monocular geometry estimation.
Dataset
Source
Publication
Num images
Storage Size
Note
NYUv2
NYU Depth Dataset V2
[1]
654
243 MB
Offical test split. Mirror, glass and window manually removed. Depth beyound 5 m truncated.
KITTI
KITTI Vision Benchmark Suite
[2, 3]
652
246 MB
Eigen's test split.
ETH3D
ETH3D SLAM & Stereo Benchmarks
[4]
454
1.3 GB
Downsized from 6202×4135 to 2048×1365
iBims-1
iBims-1 (independent… See the full description on the dataset page: https://huggingface.co/datasets/Ruicheng/monocular-geometry-evaluation.Medical_Multimodal_Evaluation_Data
Evaluation Guide
This dataset is used to evaluate medical multimodal LLMs, as used in HuatuoGPT-Vision. It includes benchmarks such as VQA-RAD, SLAKE, PathVQA, PMC-VQA, OmniMedVQA, and MMMU-Medical-Tracks.
To get started:
Download the dataset and extract the images.zip file.
Find evaluation code on our GitHub: HuatuoGPT-Vision.
This open-source release aims to simplify the evaluation of medical multimodal capabilities in large models. Please cite the relevant benchmark… See the full description on the dataset page: https://huggingface.co/datasets/FreedomIntelligence/Medical_Multimodal_Evaluation_Data.WEIRD
WEIRD
Описание задачи
WEIRD – это расширенная версия подзадачи бинарной классификации оригинального английского бенчмарка WHOOPS!. Датасет оценивает, способна ли мультимодальная модель обнаруживать нарушения здравого смысла в изображениях. Здесь нарушение здравого смысла – это ситуации, противоречащие типичным нормам реальности. Например, пингвины не могут летать, дети не водят автомобили, посетители не накладывают еду официантам, и так далее. В датасете поровну… See the full description on the dataset page: https://huggingface.co/datasets/MERA-evaluation/WEIRD.ruCLEVR
ruCLEVR
Описание задачи
RuCLEVR — это датасет для задачи визуального вопросно-ответного ризонинга (Visual Question Answering, VQA), созданный по методологии CLEVR, адаптированной для русского языка.
RuCLEVR состоит из автоматически сгенерированных изображений 3D-объектов, каждый из которых характеризуется такими признаками, как форма, размер, цвет и материал, расположенных в различных условиях и образующих сложное визуальное окружение. Набор данных включает вопросы… See the full description on the dataset page: https://huggingface.co/datasets/MERA-evaluation/ruCLEVR.ruCommonVQA
ruCommonVQA
Описание задачи
ruCommonVQA — вопросно-ответный публичный датасет на русском языке по изображениям двух типов: фото и картинки.
Вопросы делятся 1) на простые и 2) сложные, разбитые на самые частотные типы: бинарный, сравнительный, сколько, где, как, какой, что, кто, микс. Для простых вопросов нужно лишь ориентироваться на изображение, для сложных — сделать шаг ризонинга. Все изображения для сета — классические, из публичных источников, реальные фото и… See the full description on the dataset page: https://huggingface.co/datasets/MERA-evaluation/ruCommonVQA.ruNaturalScienceVQA
ruNaturalScienceVQA
Описание задачи
NaturalScienceQA представляет собой мультимодальный вопросно-ответный датасет по естественным наукам с базовыми вопросами из школьной программы, основанный на английском датасете ScienceQA. Датасет содержит вопросы по четырем дисциплинам естественных наук: физика, биология, химия и естествознание. В задании необходимо по изображению и сопроводительному контексту ответить на вопрос, выбрав правильный ответ из представленных. Задания… See the full description on the dataset page: https://huggingface.co/datasets/MERA-evaluation/ruNaturalScienceVQA.thai-ocr-evaluation
Thai OCR Evaluation Dataset
Dataset Description
The Thai OCR Evaluation Dataset is designed for evaluating Optical Character Recognition (OCR) models across various domains. It includes images and textual data derived from various open-source websites.
This dataset aims to provide a comprehensive evaluation resource for researchers and developers working on OCR systems, particularly in Thai language processing.
Data Fields
Each sample in the dataset contains… See the full description on the dataset page: https://huggingface.co/datasets/openthaigpt/thai-ocr-evaluation.Kitti-Object-Detection-Evaluation-2012Region-Evaluation-DataThis repository presents the evaluation data used for ASMv2. Please refer to this document for more details about the repository.
vlm_evaluation_v0_testThe v0 dataset is designed to evaluate the capabilities of VLMs in a non-interactive manner. This initial version primarily serves to help readers understand the structure and design of our benchmark.
Each ability dimension is represented by a dedicated directory, and within each ability directory are multiple task-specific subdirectories. For each task, there are numerous data examples. Each example includes original multi-view images, segmented images for visual prompts, a corresponding… See the full description on the dataset page: https://huggingface.co/datasets/VLABench/vlm_evaluation_v0_test.DA-2-Evaluation
DA2: Depth Anything in Any Direction
DA2 predicts dense, scale-invariant distance from a single 360° panorama in an end-to-end manner, with remarkable geometric fidelity and strong zero-shot generalization.
🎮 Usage
Please see here.
🎓 Citation
If you find these datasets useful, please consider citing 🌹:
@article{li2025depth,
title={DA$^{2}$: Depth Anything in Any Direction},
author={Li, Haodong and Zheng, Wangguangdong and He, Jing and Liu, Yuhao and… See the full description on the dataset page: https://huggingface.co/datasets/haodongli/DA-2-Evaluation.UGround-Offline-EvaluationVideo-Evaluationvtikz-evaluationface-anonymization-public-evaluation-v58
Origin Data Lab — Face Anonymization Public Evaluation V58
Privacy processing for real-world video datasets
Automated face anonymization combined with targeted Human QA for autonomous driving, robotics, computer vision, urban mobility, and AI data teams.
This repository presents a public engineering evaluation of Origin Data Lab's face anonymization pipeline in dense, low-light urban traffic conditions.
Evaluate Your Own Video
Need privacy processing for traffic… See the full description on the dataset page: https://huggingface.co/datasets/origindatalab/face-anonymization-public-evaluation-v58.BlindLoop-Evaluation
BlindLoop Evaluation
The frozen evaluation cohorts for the BlindLoop paper. Each eligible generated
task contributes exactly five deterministic, pixel-distinct image instances.
The five rows share a task's selected question/prompt family while varying the
rendered scene and gold answer as determined by the task's pixel oracle.
Config
Tasks
Rows
Documented exclusions
section1_eval5
1,298
6,490
3
section2_eval5
874
4,370
1
combined_eval5
2,172
10,860
4
Each row… See the full description on the dataset page: https://huggingface.co/datasets/taesiri/BlindLoop-Evaluation.UniScienceVQA
UniScienceVQA
Описание задачи
UniScienceVQA — мультимодальный датасет, состоящий из заданий на проверку экспертных знаний в различных областях науки (фундаментальные, социальные и прикладные науки, культуроведение, бизнес, здоровье и медицина), задания представлены в виде изображений и вопросов с аннотацией к ним. Задания делятся на три группу по варианту ответ: 1) с коротким ответом; 2) с вариантами ответов; 3) с вариантами ответов, на которые нет верного ответа.… See the full description on the dataset page: https://huggingface.co/datasets/MERA-evaluation/UniScienceVQA.SchoolScienceVQA
SchoolScienceVQA
Описание задачи
SchoolScienceVQA — русскоязычный мультимодальный датасет, созданный по методологии ScienceQA и адаптированный под школьную программу и культурный контекст России. Он содержит вопросы с выбором ответа из нескольких вариантов по дисциплинам: физика, биология, химия, экономика, история и естествознание. Каждый вопрос сопровождается изображением, текстовым контекстом и объяснением, что позволяет оценивать рассуждения модели.
Тестируемые навыки… See the full description on the dataset page: https://huggingface.co/datasets/MERA-evaluation/SchoolScienceVQA.FreiHAND_pub_v2_evaluationvlm_evaluationCapRL-Evaluation-Files
CapRL Evaluation Files
This dataset contains the files used by the CapRL Prism evaluation scripts.
Files
json_file/: 12 Prism evaluation JSON files.
bench_image_folder.zip: images used by the JSON files. After unzipping, it creates bench_image_folder/.
Each JSON stores image paths relative to the dataset root, for example:
bench_image_folder/lmm_eval_chartqa/41699051005347.png
Usage
huggingface-cli download internlm/CapRL-Evaluation-Files --repo-type dataset… See the full description on the dataset page: https://huggingface.co/datasets/internlm/CapRL-Evaluation-Files.evaluation-dataset
DeepSafe Evaluation Dataset
Evaluation set for DeepSafe,
a deepfake detection benchmark.
Tiers
Tier
Samples
Generators
Size
Use
master_eval_small/
198
116
1.7 GB
smoke test, under 2 min
master_eval/
15,454
411
10 GB
the standard benchmark
master_eval_full/
45,954
411
25 GB
complete set
Medium tier composition: 9,954 image, 3,500 audio, 2,000 video.
from huggingface_hub import snapshot_download
snapshot_download("deepsafe/evaluation-dataset"… See the full description on the dataset page: https://huggingface.co/datasets/deepsafe/evaluation-dataset.