Module texture

This module generates photorealistic surface imagery from glacier state variables (ice thickness, velocity, surface elevation, and related fields) using a pre-trained Pix2PixHD image-to-image neural network. The output is an RGB image saved as PNG or GeoTIFF at each time step, suitable for visualisation and for producing synthetic remote-sensing training data. Pre-trained model weights must be downloaded separately before use (a link is provided if the weights are not found at startup).

Parameters

Default configuration file (texture.yaml):

texture:
  format: png
  model_path: ""
  verbosity: 30
  divide_by_density: 1
  resolution: -1

Description of the parameters:

Name	Description	Default value	Units
format	Format of the texture image (png, tif, or tiff).	png	—
model_path	Name of the folder for the texture model (tf format).		—
verbosity	Python Logger verbosity level (10=DEBUG, 20=INFO, 30=WARNING, 40=ERROR, 50=CRITICAL)	30	—
divide_by_density	Divisor applied to the precipitation field when converting from model units to monthly values for texture generation; set to 1 to disable unit conversion.	1	—
resolution	Override the grid resolution for texture generation; set to -1 to use the simulation resolution.	-1	—

Contributors: Brandon Finley.