TIRAuxCloud Dataset

---

📁 Contents

Inside the hugging face repository https://huggingface.co/datasets/tirauxcloud/TIRAuxCloud/tree/main the following are provided :

• Sample data from the three test splits of the three TIRAuxCloud subsets:

  • Main Landsat
  • MA Landsat
  • VIIRS

• Saved model weights used in the experiments presented in the paper (model_files.tar.gz)

---

🔁 Reproducing Model Metrics

To reproduce the evaluation results of any model:

1️⃣ Complete the configuration file

Edit saved_model_run.json and fill in the correct configuration block for the selected subset "landsat", "landsatMA" or "viirs"

Make sure that all the parameters in the selected configuration block of the provided saved_model_run.json are filled correctly. No additional parameters are required beyond those already provided in saved_model_run.json

• "dataset_folder" parameter is the folder of the taining file lists (csv files that define which are the training, validation and test files)
• "dataset_dir" is the folder of the dataset files (image samples)
• "device", "cpuworkers" and "batch_size"** can be adjusted to your system's specifications

All the other parameters in the JSON configuration must match the corresponding values in the relevant CSV row of the selected model. All saved models are in the hugging face repository.

---

2️⃣ Run the evaluation script

Execute model_test.py with the -t parameter to select the test subset:

python model_test.py -t landsat
python model_test.py -t landsatMA
python model_test.py -t viirs