Pretrained Models

Ready-to-use geospatial AI models in ArcGIS

ArcGIS offers a growing library of over 100 pretrained models as part of Esri’s ongoing effort to make geospatial AI accessible to users of all technical backgrounds.

Aerial view of a port and shipping terminal with pretrained models detecting ships, buildings, and industrial infrastructure

What are pretrained models?

Pretrained models are deep learning models that have been trained on large volumes of data to recognize features and extract information. They come ready to use, so you don’t have to build or train a model from scratch. These models automate geospatial workflows with deep learning—a scientific AI technique—by identifying patterns and extracting information from your imagery, point clouds, video, text, time series data, and tabular data. They quickly perform tasks like feature extraction, image classification, object detection, segmentation, text analysis, and forecasting—with minimal computing resources or specialized expertise.

Why use pretrained models?

Using pretrained models provides a straightforward way to move from questions to answers quickly—without building or training models.

  • Instantly extract features from data

    Off-load tedious tasks like digitization and geographic feature extraction for roads, land cover, and building footprints with ease.

  • No need to develop or train models

    Take advantage of the models trained on huge volumes of data so you don’t have to invest time and energy into labeling datasets and model training.

  • Customizable for your specific needs

    Fine-tune or retrain some of the models using your own training data so they better match your location, landscape, resolution, or sensor type.

  • Available to all ArcGIS users

    Access a growing number of pretrained models from ArcGIS Living Atlas. We continually train and deliver new, complementary models to support your business needs. 

Previous
Next

Types of models

Pretrained models in ArcGIS come in two varieties that differ in application and flexibility.

Explore deep learning packages Compare the two types of pretrained mode

Task-specific models

Task-specific models are built to excel at one job, such as detecting building footprints, identifying swimming pools, or classifying land cover. They are highly optimized for single tasks and perform well in specific domains and imagery types. Some of these models can be fine-tuned for your specific needs.
Examples include:

Output from a pretrained land cover model mapping urban development, forests, water, and agriculture in the Great Lakes area

INTERACTIVE MAP

Explore a use case for pretrained models in ArcGIS

Powered by real data and analysis in ArcGIS Online, this web app was made with development tools from the comprehensive ArcGIS platform.

CUSTOMER SUCCESS STORIES

See how Esri customers are using pretrained models

How to access pretrained models in ArcGIS

You’ll need an ArcGIS Online or ArcGIS Enterprise license and a Professional or Professional Plus user type.

  • ArcGIS Pro

    Access pretrained models through image analysis and geoprocessing tools in ArcGIS Pro, where models are selected as part of the tool workflow.

    Explore image feature extraction
    Explore point cloud features
    Get full capabilities through Professional Plus

  • ArcGIS Online

    Run supported deep learning and analysis tools in ArcGIS Online. Models are delivered through ArcGIS Living Atlas and selected within deep learning model packages.

    Purchase a Professional or Professional Plus user type

  • ArcGIS Enterprise

    Access pretrained models or train new deep learning models through the Deep Learning Studio web application, included with ArcGIS Image Server.

    Explore ArcGIS Image Server
Previous
Next

Resources

Previous
Next

Explore deep learning packages

Browse models