--- language: - en license: other license_name: lsdetector-academic-non-commercial-license license_link: https://github.com/klaus2023/LSDetector/blob/main/LICENSE-NC.md library_name: pytorch tags: - landslide-detection - remote-sensing - semantic-segmentation - lsdetector pipeline_tag: image-segmentation --- # LSDetector Model Weights Pretrained PyTorch weights for [LSDetector](https://github.com/klaus2023/LSDetector), an open-source workflow for AI-assisted landslide mapping, target-area adaptation, expert review, and GIS-compatible inventory production. The repository provides four checkpoints from the LSDFormer and LSDSAM model families. Use them with the matching LSDetector code, model adapter, and preprocessing pipeline. ## Weights | Checkpoint | Model | Size (GB) | |---|---|---:| | `LSDFormer_LS77K.pth` | LSDFormer | 0.152 | | `LSDSAM_B_77K.pth` | LSDSAM-B | 0.441 | | `LSDSAM_H_LS77K.pth` | LSDSAM-H | 2.724 | | `LSDSAM_L_77K.pth` | LSDSAM-L | 1.359 | | **Total** | | **4.675** | Sizes use decimal units: 1 GB = 1,000,000,000 bytes. SHA-256 values are provided in `manifest.json`. ## Download Install LSDetector using the [project instructions](https://github.com/klaus2023/LSDetector#installation-and-usage-tutorial), then download the weights and manifest: ```bash hf download ZijinFu/lsdetector-weights \ LSDFormer_LS77K.pth \ LSDSAM_B_77K.pth \ LSDSAM_H_LS77K.pth \ LSDSAM_L_77K.pth \ --local-dir "$HOME/LSDetector_data_weights_v1/weights" hf download ZijinFu/lsdetector-weights manifest.json \ --local-dir "$HOME/LSDetector_data_weights_v1" ``` Register the release weights: ```bash python tools/register_release_assets.py \ --assets-root "$HOME/LSDetector_data_weights_v1" ``` ## Inference Run inference on a georeferenced GeoTIFF with a registered checkpoint: ```bash lsd deploy \ --registry-root "$PWD" \ --weight-id LSDFormer_LS77K \ --geotiff /path/to/input.tif \ --output-dir "$HOME/LSDetector-runs/first-map" \ --device auto ``` Use `LSDSAM_B_77K`, `LSDSAM_L_77K`, or `LSDSAM_H_LS77K` for the corresponding LSDSAM variants. See the [LSDetector CLI guide](https://github.com/klaus2023/LSDetector/blob/main/docs/cli_usage.md) for configuration and preprocessing details. ## License and citation These weights are released for academic research, education, and other non-commercial scientific use under the [LSDetector Academic Non-commercial License](https://github.com/klaus2023/LSDetector/blob/main/LICENSE-NC.md). Commercial use requires written permission. Cite the corresponding [LSDFormer](https://doi.org/10.1007/s10346-026-02733-w) or [LSDSAM](https://doi.org/10.1109/JSTARS.2026.3678141) paper when using these weights.