Tools tagged with "Labeled Datasets"

GeeFlow is a Python library from Google DeepMind designed for generating and processing large-scale geospatial datasets using Google Earth Engine. It focuses on supporting geospatial AI research by creating datasets in TFDS format, ready for model training, and enabling the creation of inference maps up to global scale.

CDrone introduces the CARLA Drone dataset, expanding camera perspectives for monocular 3D object detection benchmarks. It also develops GroundMix, an effective data augmentation pipeline that significantly boosts detection accuracy. This work aims to encourage progress in 3D detection frameworks.

This dataset, hosted by the ETH Zurich Research Collection, provides a global canopy height map for 2020. It was created using data from Sentinel-2 and GEDI, offering valuable insights into forest structures and biomass. The resource is open access and available under a Creative Commons license.

LILA BC provides a comprehensive list of conservation-related labeled datasets that are not directly hosted on the LILA platform. This resource complements LILA's own dataset collection, aiming to track machine-learning-ready data with interesting labels attached to images or documents. The page emphasizes publicly accessible datasets relevant to various conservation efforts.

The Computer Vision Group at TUM offers a variety of datasets for researchers in computer vision, image processing, and pattern recognition. These datasets are shared to foster academic collaboration, with a request to cite the relevant publications when utilized.

This repository provides an ensemble of AI models, known as SpeciesNet, developed by Google to classify wildlife in camera trap images. It combines an object detector (MegaDetector) to find objects of interest and an image classifier to identify species, significantly accelerating wildlife monitoring efforts. These models are utilized within the Wildlife Insights platform.

LILA BC provides a curated collection of publicly available datasets, primarily consisting of camera trap and underwater video imagery. These datasets support the development and training of machine learning models for wildlife monitoring, biodiversity analysis, and environmental impact studies. The platform aggregates data from various conservation projects and agencies worldwide.

The National Archive of Criminal Justice Data (NACJD) serves as a vital resource for research in criminal justice and criminology. It provides comprehensive data services, enabling users to both discover and share valuable datasets. This initiative supports empirical research and informs policy development within the criminal justice field.

Satlas Super Resolution offers globally generated AI-enhanced imagery for 2023, aiming for high accuracy and frequent updates. This GitHub repository provides the training and inference code, data, and model weights for the super-resolution outputs. It supports the mission of Satlas to provide open, AI-generated geospatial data.

This dataset combines airborne lidar and Landsat satellite data to map vegetation structure in Mexican ecosystems. It includes lidar-derived canopy height values and Landsat-derived vegetation indices for 499,925 sample points across 1,105 aerial images, enabling fine-resolution ecological restoration monitoring.

SensatUrban is a large-scale photogrammetric point cloud dataset designed for urban semantic segmentation research. It features nearly three billion richly annotated points, covering approximately 6 km² across two UK cities. This dataset provides essential benchmarks and challenges for advancing 3D point cloud analysis and deep learning applications.

The KITTI Vision Benchmark Suite offers challenging real-world datasets for computer vision tasks in autonomous driving. Data is collected using a research vehicle equipped with high-resolution cameras, a Velodyne laser scanner, and a GPS system in various environments around Karlsruhe, Germany. It provides benchmarks for tasks like stereo, optical flow, visual odometry, 3D object detection, and 3D tracking.

This GitHub repository provides a curated list of ground-truth forest datasets for the machine learning and forestry community. It focuses on in-situ and validation data for various analyses, including biodiversity, carbon, and wildfire. The goal is to help researchers and practitioners easily find datasets for building machine learning models for forest analysis.

FlexRoad is a framework that transforms noisy 3D road reconstructions into smooth, accurate, and realistic road meshes. It addresses limitations of existing methods by fitting Non-Uniform Rational B-Splines (NURBS) surfaces to 3D road points. The framework is crucial for autonomous driving, urban planning, and virtual simulations, ensuring high-quality road models by reducing artifacts and inconsistencies found in typical planar representations.

RarePlanes is an open-source machine learning dataset from CosmiQ Works and AI.Reverie, designed to test the value of synthetic data in computer vision. It combines real Maxar WorldView-3 satellite imagery with synthetically generated data to enhance algorithms for automatic aircraft detection and attribute identification. This dataset is the largest openly available very high-resolution resource for synthetic data testing in an overhead perspective.

The FloodPlanet dataset provides manually labeled inundation data, meticulously collected for advancing flood detection research. It serves as a crucial resource for training and validating deep learning models, utilizing high-resolution imagery from PlanetScope, Sentinel-1, and Sentinel-2 satellites. This version 1.0 dataset, contributed by Zhijie(JJ) Zhang and collaborators, is a valuable asset for the geospatial community.

The TACO dataset addresses the critical need for annotated images to train deep learning models for trash detection. It aims to advance AI applications in environmental monitoring, such as drone-based trash surveying and robotic litter collection, by providing a comprehensive and continually updated collection of segmented trash images.