Skip to main content

Toulouse Hyperspectral Dataset evaluation

The Toulouse Hyperspectral Data Set is a public land-cover dataset with standard train and test splits. It can be used to evaluate a Clarity Classification model against a reproducible external protocol.

This repository includes a utility for comparing Toulouse runs, but it does not contain a checked-in run manifest or exported metric artifacts that verify a Clarity benchmark score. For that reason, this page does not publish accuracy or F1 claims for the example shown below.

Reproducible evaluation​

Before reporting a result, record:

  • the official dataset release and split identifier;
  • the exact Clarity dataset and model version IDs;
  • the source bands, center wavelengths, and processing level;
  • preprocessing, augmentation, model initialization, and hyperparameters;
  • the Clarity code version recorded on the trained model;
  • the evaluation unit and treatment of unlabeled pixels; and
  • the metric aggregation method across classes and splits.

Use the official spatial splits without moving test pixels into training or validation. Report overall accuracy together with macro and per-class precision, recall, F1, and support. Aggregate multiple splits consistently and include variation across them.

Example visualization​

Example Clarity inference visualization for a Toulouse test tile. The left pane is a color composite and the right pane is a categorical model output.

Example inference visualization. The image is qualitative and is not evidence of a benchmark score.

Inspect output at full resolution for class confusion, boundary effects, spatial leakage, and systematic errors. A visually plausible map does not replace held-out metrics.

Dataset reference​

Thoreau, R., Risser, L., Achard, V., Berthelot, B., and Briottet, X. (2024). “Toulouse Hyperspectral Data Set: A benchmark data set to assess semi-supervised spectral representation learning and pixel-wise classification techniques.” ISPRS Journal of Photogrammetry and Remote Sensing, 212, 323–337.

Dataset website: Toulouse Hyperspectral Data Set