File size: 2,712 Bytes
c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 8507641 c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa c3e2316 bd894aa | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | ---
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.
|