95-Cloud

Dataset Overview

Dataset Name

• 95‑Cloud

Dataset Description

• 95‑Cloud is an extended cloud‑detection dataset built upon the previously released 38‑Cloud dataset.

Dataset Content

• Training Set: 34,701 patches of 384 × 384 pixels, mainly extracted from 75 Landsat 8 Collection 1 Level‑1 scenes covering North America.
• Test Set: Identical to the 38‑Cloud test set, containing 9,201 patches from 20 scenes.

Dataset Characteristics

• The training set adds 57 new scenes compared with 38‑Cloud.
• Each patch provides four corresponding spectral bands: red (band 4), green (band 3), blue (band 2), and near‑infrared (band 5), stored in their respective directories.
• The directory structure mirrors that of 38‑Cloud.

Download and Usage

• The dataset is split into two download parts:
  1. The 95‑Cloud training set covering 57 scenes.
  2. The remaining training scenes (38 scenes) and the 20 test scenes.
• After downloading, combine the parts.

Notes

• Thin clouds and haze are treated as cloud (same as thick cloud).
• Natural‑color images are pseudo‑color visualizations for inspection only and are not used for training or testing.
• A CSV file listing 21,502 informative patches (with less than 80 % black borders) is provided.

Citation

• If you use this dataset, please cite the associated paper.