Esri/deep-learning-frameworks

Installation support for Deep Learning Frameworks for the ArcGIS System

What it solves

This project provides a comprehensive installer for the deep learning and machine learning libraries required to run AI tools within the ArcGIS ecosystem (ArcGIS Pro, Server, and the ArcGIS API for Python). It eliminates the manual effort of configuring complex environments by bundling 254 packages—including PyTorch, transformers, Fast.ai, and scikit-learn—ensuring compatibility between the AI frameworks and the geospatial software.

How it works

The installer adds a curated set of deep learning packages to the default arcgispro-py3 Python environment. It supports multiple platforms, including Windows (via .msi installers) and Linux (via .sh scripts). For users in disconnected environments, it provides separate "backbone" packages to allow model training without internet access. The tools are designed to leverage NVIDIA GPUs via CUDA for accelerated computation, though some components can fall back to the CPU.

Who it’s for

Geospatial analysts and developers using ArcGIS who need to perform AI tasks such as feature extraction, pixel classification, and feature categorization on geographic data.

Highlights

  • Extensive Library Bundle: Includes 254 packages, including industry standards like PyTorch and scikit-learn.
  • Geospatial Integration: Directly enables the arcgis.learn module and integrated Geoprocessing tools for object detection and training.
  • Cross-Platform Support: Dedicated installers for ArcGIS Pro (Windows), ArcGIS Server (Windows/Linux), Drone2Map, and Reality Studio.
  • Offline Capability: Offers specific backbone packages for air-gapped or disconnected environments.

Related

  • Project
    opengeos/geoai

    A Python package that integrates AI frameworks with geospatial data analysis, providing tools for satellite imagery processing, model training, and inference.

  • Project
    neuralhydrology/neuralhydrology

    A PyTorch-based Python library designed for deep learning research in hydrology, allowing users to train neural networks via configuration files.

  • Project
    huggingface/pytorch-image-models

    timm (PyTorch Image Models) is a library that provides hundreds of pre‑trained vision architectures (CNNs, Vision Transformers, hybrids, etc.) together with training scripts, optimizers, schedulers, data utilities, and hyper‑parameter sweep tools, all built on PyTorch for easy experimentation and deployment.

  • Project
    frgfm/holocron

    A library of high-quality PyTorch implementations of recent deep learning tricks and model architectures for computer vision, including classification, detection, and segmentation.

  • Project
    kozistr/pytorch_optimizer

    A production-focused optimization toolkit for PyTorch providing over 100 optimizers, schedulers, and loss functions through a consistent API.