User guide

The main objective of orthoseg is to make it easier to automatically digitize features on orthophotos. This manual will guide you through the process of setting up a new project, creating a training dataset, training a neural network and running a prediction on an orthophoto.

Smaller projects are no problem to run on a regular laptop or desktop computer. Only if you want to use more complex models, train on larger training datasets and/or want to process large areas using a decent CUDA GPU will become very recommended to reduce waiting times.

Contents

• 1. Run sample project
• 2. Prepare a new project
  • 2.1. Only once: prepare “projects” directory
  • 2.2. Add the layer(s) you want to segment on to the image layer configuration
  • 2.3. Project name
  • 2.4. Prepare “project” directory
  • 2.5. Prepare project settings
  • 2.6. Configure image layer(s)
  • 2.7. Prepare training data files
  • 2.8. Prepare GIS project
• 3. Finetune your project
  • 3.1. What model architecture should I use?
• 4. Create training data
  • 4.1. General info
  • 4.2. Search for good examples and add them to the training input
  • 4.3. Run training + (optionally) a (full) prediction
• 5. Run train, predict, session