RTS-dataset

This dataset accompanies the SA-SegFormer RTS code repository for binary semantic segmentation of retrogressive thaw slumps (RTS). It contains raster chips, segmentation masks, multi-source environmental factor rasters, and CSV manifests used for training, validation, testing, and temporal-transfer experiments. The archive includes: - 2023 dataset: 1195 complete chips.- Required main split for model training: 837 training chips, 179 validation chips, and 179 testing chips.- 2024 dataset: 717 complete chips for cross-year/temporal-transfer experiments.- Three-fold test/evaluation outputs and manifests.- CSV manifests mapping each chip to its optical image, binary mask, and factor rasters. Each SA-SegFormer sample is constructed from 13 input channels: RGB optical imagery plus 10 factor rasters: DEM, EVI, FTI, LST, NBR, NDMI, NDVI, TCB, TCG, and TCW. Original Chinese region names and tile filenames are preserved to maintain correspondence with the manifests. The accompanying code is available at: https://github.com/FX794726198/SA-SegFormer-RTS-open After downloading, unzip the dataset archive at the root of the code repository and run: ```bashpython scripts/summarize_dataset.py