s2-100k-preprocessed

S2-100K Preprocessed Derived from torchgeo/s2-100k. What changed The original dataset stores patches as uint16 GeoTIFF files inside plain tar archives with no compression. This version converts every patch to a blosc2/zstd-compressed float32 array for faster I/O in training pipelines. No season selection is performed. Each patch in the source dataset is a single Sentinel-2 L2A acquisition (no temporal dimension), so every patch is included as-is.… See the full description on the dataset page:

Source: Hugging Face Hub (moham741/s2-100k-preprocessed). Metadata imported from the dataset’s Hub tags.