BiliSakura / CUT-OpenEarthMap-SAR

CUT-OpenEarthMap-SAR

CUT (Contrastive Unpaired Translation) models for SAR ↔ optical image translation, converted for use with cut_lib and pytorch-image-translation-models .

Models

## Usage with pytorch-image-translation-models

Use the community example at `examples/community/openearthmap_sar.py`:

```python
from examples.community.openearthmap_sar import load_openearthmap_sar_pipeline

# SAR → Optical
pipeline = load_openearthmap_sar_pipeline(
    checkpoint_dir="/root/worksapce/models/BiliSakura/OpenEarthMap-SAR",
    model_name="sar2opt",
    device="cuda",
)
output = pipeline(source_image=pil_image, output_type="pil")

CLI (from projects/pytorch-image-translation-models with conda activate rsgen ):

python -m examples.community.openearthmap_sar --input sar.png --output out.png

Note: Pass source_image as PIL.Image (not raw numpy). The module provides OpenEarthMapSARGenerator with anti-aliased down/upsampling to match the original CUT training.

Safetensors format

Each model can be converted from .pth to safetensors + config.json for safer loading:

cd projects/pytorch-image-translation-models
conda activate rsgen
python scripts/convert_openearthmap_sar_to_safetensors.py \
    --checkpoint-dir /root/worksapce/models/BiliSakura/OpenEarthMap-SAR

This creates per-model layout aligned with CUTPipeline and train_cut.py :

{model_name}/
  generator/
    config.json
    diffusion_pytorch_model.safetensors

load_openearthmap_sar_pipeline prefers safetensors when present; otherwise falls back to .pth .

Architecture

• in_channels : 3
• out_channels : 3
• base_filters : 64
• n_blocks : 9
• norm : instance

Reference

@ARTICLE{11303033,
  author={Xia, Junshi and Chen, Hongruixuan and Broni-Bediako, Clifford and Wei, Yimin and Song, Jian and Yokoya, Naoto},
  journal={IEEE Geoscience and Remote Sensing Magazine}, 
  title={OpenEarthMap-SAR: A benchmark synthetic aperture radar dataset for global high-resolution land cover mapping [Software and Data Sets]}, 
  year={2025},
  volume={13},
  number={4},
  pages={476-487},
  keywords={Translation;Semantic segmentation;Source coding;Urban planning;Land surface;Geoscience and remote sensing;Benchmark testing;Software;Sustainable development;Synthetic aperture radar},
  doi={10.1109/MGRS.2025.3599512}}