torchgeo/torchgeo

Features

• Remote Sensing Machine Learning - Provides a framework for building and training PyTorch models to analyze satellite and aerial imagery.
• Image Data Preprocessing - Provides tools for cleaning and transforming raw raster and vector data for neural network pipelines.
• Pre-training Transfer Learning - Provides extensive support for leveraging pre-trained weights from diverse sensors to train models on small datasets.
• Remote Sensing Transfer Learning - Adapts models pre-trained on global satellite imagery to perform specific geospatial tasks with limited labeled data.
• Geospatial Data Modules - Wraps geospatial loading and sampling logic into standard PyTorch tensor pipelines for seamless deep learning training.
• Multispectral Band Mapping - Maps pre-trained neural network weights from specific satellite sensor bands to custom multispectral input channels.
• Geospatial Machine Learning - Serves as a framework for building and training neural networks for semantic segmentation, object detection, and change detection.
• Geospatial Development Libraries - Implements a PyTorch-based library for deep learning on geospatial data, handling remote sensing imagery and coordinate systems.
• Coordinate-Based Spatial Sampling - Extracts small image patches from large-scale geospatial datasets using coordinate-based samplers.
• Coordinate Reference System Transformations - Provides automatic reprojection of multispectral imagery between different coordinate reference systems during data loading.
• Imagery and Mask Loaders - Enables loading of uncurated raster imagery and vector masks for custom model training and inference.
• Raster Data Loaders - Implements specialized loading for raster files incorporating custom coordinate reference systems and band selection.
• Training Data Pipelines - Manages the pipeline for splitting, sampling, and loading geospatial data into mini-batches for training.
• Geospatial Dataset Management - Manages the loading and sampling of large-scale geographic datasets with automatic coordinate reference system and resolution handling.
• Geospatial Sampling Strategies - Provides specialized sampling strategies to extract spatial samples and batches from geospatial data sources.
• Grid-Based Spatial Sampling - Extracts chips in a structured grid pattern with configurable stride to cover specific regions of interest.
• Remote Sensing Dataset Loaders - Provides specialized tools for loading, reprojecting, and sampling raster and vector data from satellite and aerial sensors.
• Geospatial Metadata Extractors - Handles imagery with geographic metadata, including automatic coordinate reprojection and resolution matching.
• Multispectral Processors - Provides a suite of spatial transformations and indices for preparing multispectral satellite imagery.
• Multispectral Image Enhancements - Calculates normalized difference indices and applies spatial rearrangements to enhance multispectral satellite imagery.
• Spatial-Level Transformations - Implements specialized spatial transformations on satellite and aerial imagery to prepare data for neural networks.
• Spatial - Extracts random chips from a region of interest to maximize dataset variety during training.
• Geospatial Dataset Splitters - Divides datasets using random bounding box assignment, grid-cell sampling, or temporal splitting to prevent spatial leakage.
• End-to-End Training Pipelines - Provides integrated workflows for executing end-to-end geospatial machine learning experiments.
• Geospatial Model Architectures - Implements diverse model architectures to process time series, 2D images, 3D change detection, and 4D atmosphere data.
• Geospatial Pre-training Datasets - Provides support for loading large-scale global imagery collections designed for foundation model development.
• Pre-trained Model Zoos - Ships a dedicated collection of pre-trained weights and architectures optimized for geospatial tasks.
• Pre-trained Model Application - Enables the use of specialized pre-trained weights for segmentation and object detection on geographic data.
• Remote Sensing Task Workflows - Manages training and evaluation workflows for specialized remote sensing tasks such as change detection and semantic segmentation.
• Satellite Datasets - Utilizes pre-trained weights and specialized architectures to extract insights from multispectral and temporal geographic data.
• Geospatial Dataset Joins - Joins disparate geospatial datasets by aligning them based on overlapping geographic bounding boxes.
• Geospatial Dataset Unions - Combines multiple image sources or labels using union operations to create multimodal or spatially aligned datasets.
• Benchmark Dataset Loaders - Loads standardized geospatial benchmark datasets containing images and target labels for machine learning tasks.
• Vector Rasterizers - Converts geospatial vector masks into pixel-based grids to create ground truth labels for semantic segmentation.
• Device-Side Tensor Transformations - Implements spatial augmentations and normalized difference indices directly on GPU tensors for efficient deep learning pipelines.
• Geospatial Dataset Splitting - Divides geospatial datasets using geographic region boundaries to ensure a rigorous separation between training and validation sets.

Historial de estrellas