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.