datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Sea-Undistort
Dataset Card for Sea-Undistort
Sea-Undistort is a synthetic dataset for through-water image restoration in high-resolution airborne bathymetry. It contains 1,200 scenes with four 512×512 RGB images per scene: (1) ground/no water, (2) undistorted/no waves, (3) no sunglint, (4) distorted (all effects). Each scene comes with structured per-image metadata describing camera, water, sky/illumination, and seafloor parameters. Images were procedurally rendered in Blender to emulate… See the full description on the dataset page: https://huggingface.co/datasets/maxkromer/Sea-Undistort.FUSU-Fine_grained_Urban_Semantic_Understanding
About:
FUSU dataset covers 5 whole urban areas, 847 km^2 located in the north and south of China, with 17 land use and land cover (LULC) classes and over 170K images and 30 billion pixels of annotations, supporting segmentation, change detection and domain adaptation tasks. This data comprises 2 parts:
Bi-temporal high-resolution satellite RGB images with fine-grained annotations.
Monthly revisited Sentinel-2 and Sentinel-1 images.
Details:
1.… See the full description on the dataset page: https://huggingface.co/datasets/sp-juni/FUSU-Fine_grained_Urban_Semantic_Understanding.reef-guidance-system
Dataset Card for Reef Guidance System
This dataset provides imagery used for training and evaluation of models in the Reef Guidance System. All imagery was collected by the Australian Institute of Marine Science using the ReefScan™ Transom Marine Monitoring System.
If you use this dataset in your work, please cite the associated paper: AI-driven dispensing of coral reseeding devices for broad-scale restoration of the Great Barrier Reef (citations provided at bottom of this… See the full description on the dataset page: https://huggingface.co/datasets/QCR-Underwater-Perception/reef-guidance-system.IMAGE_UNDERSTANDINGA key question for understanding multimodal performance is analyzing the ability for a model to have basic
vs. detailed understanding of images. These capabilities are needed for models to be used in
real-world tasks, such as an assistant in the physical world. While there are many dataset for object detection
and recognition, there are few that test spatial reasoning and other more targeted task such as visual prompting.
The datasets that do exist are static and publicly available, thus… See the full description on the dataset page: https://huggingface.co/datasets/microsoft/IMAGE_UNDERSTANDING.kitchen-workspace-understanding-safe-manipulation
Kitchen Workspace Understanding & Safe Manipulation
Generated by datapack-import.ts
This dataset mirrors public data-pack render outputs from Physicl.
Each row represents one render view. The image column contains a stable URL to the primary render image uploaded under /data; image_path stores the relative repository path and data_commit_sha pins the Hugging Face dataset commit used by those URLs. Files are uploaded as downloaded unless optional PNG recompression is enabled by… See the full description on the dataset page: https://huggingface.co/datasets/physicl/kitchen-workspace-understanding-safe-manipulation.atlas-31-strengthening-candidate-verification-under-rl
31. Strengthening candidate verification under reinforcement learning
1. Question and links
Read this first. The reading copy of this directory is t2ance/atlas-experiments under 31-strengthening-candidate-verification-under-rl/; the saved training steps and the per-token training arrays are on the Hugging Face repository t2ance/atlas-31-strengthening-candidate-verification-under-rl only.
How can reinforcement learning make the orchestrator's comparing and… See the full description on the dataset page: https://huggingface.co/datasets/t2ance/atlas-31-strengthening-candidate-verification-under-rl.casia-char-1
CASIA Character Sample Dataset
This dataset is adapted from CASIA Online and Offline Chinese Handwriting Databases,
but this only contains character level sample data (from the offline database). The first column is the ground truth label (single character from
GB2312 charset) and the second one is byte sequences of the decoded PNG files from the original .gnt files.
Conditions of Academic Use
Please refer to the official page for more information.
All samples in the… See the full description on the dataset page: https://huggingface.co/datasets/UndefinedCpp/casia-char-1.cc12m-wds-coco-recaptioned
CC12M WebDataset with COCO-style Recaptions
A large-scale image-text dataset containing 3 million images from Conceptual Captions 12M (CC12M) with COCO-style factual descriptions generated using NVIDIA Nemotron Nano 12B v2 VL.
Dataset Overview
Base Dataset: pixparse/cc12m-wds - Conceptual Captions 12M (CC12M)
Images: 3,000,000+ high-quality internet images
Recaption Model: NVIDIA Nemotron Nano 12B v2 VL
Recaption Style: COCO-style factual descriptions (20 words average)… See the full description on the dataset page: https://huggingface.co/datasets/undefined443/cc12m-wds-coco-recaptioned.form_understanding_in_noisy_scanned_documents_plus
Dataset Card for Form Understanding in Noisy Scanned Documents Plus
This is a FiftyOne dataset with 1026 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/form_understanding_in_noisy_scanned_documents_plus")
# Launch the App… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/form_understanding_in_noisy_scanned_documents_plus.MUOT_3M-A_3_Million_Frame_Underwater_Object_Tracking_Dataset
🌊 MUOT-3M: The Largest Multimodal Underwater Object Tracking Dataset
Official repository for MUOT-3M📄 MUOT-3M: The Largest Multimodal Underwater Object Tracking Dataset and MUTrack Tracking Method
🚀 Overview
MUOT-3M is currently the largest underwater object tracking dataset, containing over 3 million annotated frames across 3,030 underwater videos with synchronized multimodal annotations.
The benchmark is designed to advance research in:
Underwater object tracking… See the full description on the dataset page: https://huggingface.co/datasets/AhsanBB/MUOT_3M-A_3_Million_Frame_Underwater_Object_Tracking_Dataset.underwater-objects-5v7p8
Underwater Objects 5V7P8
This dataset is part of the Roboflow 100 benchmark, a diverse collection of 100 object detection datasets spanning 7 imagery domains.
Dataset Statistics
Split
Images
Train
5,320
Validation
1,520
Test
760
Total
7,600
Classes (5)
echinus
holothurian
scallop
starfish
waterweeds
Usage
With LibreYOLO
from libreyolo import LIBREYOLO
# Load a model
model = LIBREYOLO(model_path="libreyoloXnano.pt")… See the full description on the dataset page: https://huggingface.co/datasets/LibreYOLO/underwater-objects-5v7p8.multi-view-bathroom-scene-understanding-camera-relocalization
Multi-View Bathroom Scene Understanding & Camera Relocalization
Generated by datapack-import.ts
This dataset mirrors public data-pack render outputs from Physicl.
Each row represents one render view. The image column contains a stable URL to the primary render image uploaded under /data; image_path stores the relative repository path and data_commit_sha pins the Hugging Face dataset commit used by those URLs. Files are uploaded as downloaded unless optional PNG recompression is… See the full description on the dataset page: https://huggingface.co/datasets/physicl/multi-view-bathroom-scene-understanding-camera-relocalization.synthetic-code-understanding
SYNTHETIC-1
This is a subset of the task data used to construct SYNTHETIC-1. You can find the full collection here
sentence-undl_zh2en_alignedraw:bot-yaya/undl_zh2en_aligned
work:split
underninja
Bangumi Image Base of Under Ninja
This is the image base of bangumi Under Ninja, we detected 38 characters, 3978 images in total. The full dataset is here.
Please note that these image bases are not guaranteed to be 100% cleaned, they may be noisy actual. If you intend to manually train models using this dataset, we recommend performing necessary preprocessing on the downloaded dataset to eliminate potential noisy samples (approximately 1% probability).
Here is the characters'… See the full description on the dataset page: https://huggingface.co/datasets/BangumiBase/underninja.physical-ai-bench-understanding
Physical AI Bench - Understanding
PAI-Bench (Physical AI Bench) is a comprehensive benchmark designed to evaluate physical AI generation and understanding capabilities across various real-world scenarios. This particular dataset, PAI-Bench-U, focuses specifically on Video Understanding tasks, comprising 2,808 real-world cases with task-aligned metrics.
Paper: PAI-Bench: A Comprehensive Benchmark For Physical AI
Code: GitHub Repository
Citation
If you use Physical AI… See the full description on the dataset page: https://huggingface.co/datasets/shi-labs/physical-ai-bench-understanding.underwater-datasets
Updates / News
April 28, 2025:We are uploading new datasets collected in the Trondheim Fjord and at the Marine Cybernetics Lab Pool.Additionally, we are migrating the dataset repository from github.com/ntnu-arl/underwater-datasets to huggingface.co/datasets/ntnu-arl/underwater-datasets.
Quick Start
You can quickly download the dataset using the huggingface_hub Python library.
1. Install huggingface_hub
pip install huggingface_hub
2.… See the full description on the dataset page: https://huggingface.co/datasets/ntnu-arl/underwater-datasets.Draw-and-Understand
🎨 Draw-and-Understand: Leveraging Visual Prompts to Enable MLLMs to Comprehend What You Want
The interaction between humans and artificial intelligence (AI) is a crucial factor that reflects the effectiveness of multimodal large language models (MLLMs). However, current MLLMs primarily focus on image-level comprehension and limit interaction to textual instructions, thereby constraining their flexibility in usage and depth of response. Therefore, we introduce the… See the full description on the dataset page: https://huggingface.co/datasets/Afeng-x/Draw-and-Understand.underwoodunderwater-pipes-4ng4t
underwater pipes > release-640
https://universe.roboflow.com/roboflow-100/underwater-pipes-4ng4t
This dataset is part of RF100, an Intel-sponsored initiative to create a new object detection benchmark for model generalizability.
Dataset Summary
Total images: 7971
Train: 5617 images
Validation: 1575 images
Test: 779 images
Classes: 1 (pipe)
Format: YOLOv8 (Ultralytics)
License: CC BY 4.0
Preprocessing
Auto-orientation of pixel data (with EXIF-orientation… See the full description on the dataset page: https://huggingface.co/datasets/LibreYOLO/underwater-pipes-4ng4t.Thai-understanding
Thai-Understanding: Thai-SUP & XLSR-Thai
Overview
Thai-Understanding is an open-source repository that provides a solution for speech understanding in the Thai language. This repository includes:
Thai-SUP: The first open-source Thai speech understanding dataset, which includes over 1,000 hours of data across three tasks: Intent Classification (IC), Named Entity Recognition (NER), and Speech Rephrasing (SR).
XLSR-Thai: The first large-scale self-supervised learning (SSL)… See the full description on the dataset page: https://huggingface.co/datasets/mcshao/Thai-understanding.varroa-yolo-under-2m-wiouundefinedAgentic-Long-Context-Understanding-QA 📖 Agentic Long Context Understanding 📖
Self-Taught Agentic Long Context Understanding (Arxiv).
AgenticLU refines complex, long-context queries through self-clarifications and contextual grounding, enabling robust long-document understanding in a single pass.
Installation Requirements
This codebase is largely based on OpenRLHF and Helmet, kudos to them.
The requirements are the same
pip install openrlhf
pip install -r ./HELMET/requirements.txt… See the full description on the dataset page: https://huggingface.co/datasets/yzhuang/Agentic-Long-Context-Understanding-QA.Underwater-Acoustic-Channel-Repository
Underwater Acoustic Channel Repository
This Hugging Face dataset is a structured, checksum-preserving mirror of version 1.0 of the Underwater Acoustic Channel Repository. The original dataset was published by Zhengnan Li, Mandar Chitre, Diego Cuji, James Preisig, Andrew Singer, Milica Stojanovic, and Paul van Walree.
The collection contains measured underwater acoustic channel impulse responses (CIRs) from eight at-sea experimental groups. Channel and accompanying noise files… See the full description on the dataset page: https://huggingface.co/datasets/UWA-CP/Underwater-Acoustic-Channel-Repository.rework_undl_textPlease, visit our Github repo at v4dev branch for up-to-date deploying notes and reproducing details. https://github.com/mnbvc-parallel-corpus-team/UPRPRC/tree/v4dev
The main branch will stay unchanged until review process is done.
2025/12/03 Updated
Hello everyone, please download the updated and more complete file-level alignment dataset: https://huggingface.co/datasets/bot-yaya/UPRPRC_FTXT_FILEWISE
Or, if you can't feed in the entire file, download the paragraph-level alignment… See the full description on the dataset page: https://huggingface.co/datasets/bot-yaya/rework_undl_text.VRI_Underwater_GraspingMulti-Turn-Insurance-Underwriting
Dataset Card for Multi-Turn-Insurance-Underwriting
Dataset Summary
This dataset includes sample traces and associated metadata from multi-turn interactions between a commercial underwriter and AI assistant. We built the system in langgraph with model context protocol and ReAct agents. In each sample, the underwriter has a specific task to solve related to a recent application for insurance by a small business. We created a diverse sample dataset covering 6 distinct types… See the full description on the dataset page: https://huggingface.co/datasets/snorkelai/Multi-Turn-Insurance-Underwriting.pixmo-point-count-gen-undmc4_und_idfiltered,deduplication MC4-ID from MC4 part undfined