# CondDiff · SAR2EO

CondDiff (GRSL'23) retrained by the GeoSET authors on the SAR2EO split. These are the weights behind
the CondDiff row for SAR2EO in Table 3 of the GeoSET paper
([arXiv:2609.37496](https://arxiv.org/abs/2609.37496)).

- Method: Bai, Pu and Xu, "Conditional diffusion for SAR to optical image translation", GRSL 2023.
- Implementation: [Coordi777/Conditional-Diffusion-for-SAR-to-Optical-Image-Translation](https://github.com/Coordi777/Conditional-Diffusion-for-SAR-to-Optical-Image-Translation) at commit `8326f54adf3a8d717792cb7abc7d3d2b7dfac565`.
- Training and inference code: [`baselines/conddiff`](https://github.com/KAIST-VICLab/GeoSET/tree/main/baselines/conddiff) in the GeoSET repository.

## Files

| File | Bytes | SHA-256 | Role |
|---|---:|---|---|
| `ema_final.pt` | 657,495,287 | `3fda16a002d04ce90156f8d101a5db2337c111c683c8c2673f443268800e5c3d` | UNet EMA (0.9999) weights at step 50,000 (256 × 256 model) |

## Results (GeoSET paper, Table 3)

SAR2EO test split: the first 4,000 of the 21,260 pairs of the E3Diff test split, 256 × 256.

| FID↓ | DISTS↓ | KID↓ | DINO↑ | LPIPS↓ | SSIM↑ | PSNR↑ |
|---:|---:|---:|---:|---:|---:|---:|
| 113.2 | 0.317 | 0.0639 | 0.461 | 0.551 | 0.334 | 11.12 |

## Inference

```bash
git clone https://github.com/KAIST-VICLab/GeoSET && cd GeoSET
export BASELINES_ROOT=/path/to/upstream_code DATA_ROOT=/path/to/data WORK_DIR=/path/to/work CKPT_ROOT=/path/to/checkpoints
bash baselines/conddiff/fetch.sh
python baselines/data/diff_data.py conddiff sar2eo
hf download JeonghyeokDo/GeoSET --include "baselines/sar2eo/conddiff/*" --local-dir hf
mkdir -p $CKPT_ROOT/conddiff/sar2eo && cp hf/baselines/sar2eo/conddiff/ema_final.pt $CKPT_ROOT/conddiff/sar2eo/
bash baselines/conddiff/test.sh sar2eo        # writes $WORK_DIR/results/conddiff/sar2eo/<stem>.png
python scripts/evaluate.py --dataset sar2eo --pred-dir $WORK_DIR/results/conddiff/sar2eo
```

## Licence

The CondDiff repository contains no licence file; it is derived from [openai/guided-diffusion](https://github.com/openai/guided-diffusion) (MIT). These weights were trained by the GeoSET authors with that code; the GeoSET repository distributes only its changes to the code.