baselines/sar2opt/e3diff/README.md

E3Diff · SAR2Opt

E3Diff (GRSL'24) retrained by the GeoSET authors on the SAR2Opt split. These are the weights behind the E3Diff row for SAR2Opt in Table 2 of the GeoSET paper (arXiv:2609.37496).

• Method: Qin, Zou, Li and Zhang, "Efficient end-to-end diffusion model for one-step SAR-to-optical translation", GRSL 2024.
• Implementation: DeepSARRS/E3Diff at commit 38601093ab8f8e4b478144621f20890b100a3b74 (stage 2, initialised from the DDPM (SR3) stage-1 weights).
• Training and inference code: baselines/e3diff in the GeoSET repository.

Files

File	Bytes	SHA-256	Role
gen.pth	768,426,944	565c45c53434666fa0f62dbdbfa772742ae8b72eb204b324b05953bdf246d623	One-step generator UNet (E3Diff stage 2), state dict at iteration 310,000; 3-channel output, [SAR-PPB, SAR-canny, SAR] condition

Results (GeoSET paper, Table 2)

SAR2Opt test split, 627 pairs, centre 512 × 512 crop.

FID↓	DISTS↓	KID↓	DINO↑	LPIPS↓	SSIM↑	PSNR↑
104.7	0.232	0.0306	0.541	0.529	0.249	16.09

Inference

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/e3diff/fetch.sh
# vision-aided-loss (CLIP discriminator, built by the model): see baselines/e3diff/README.md
python baselines/data/diff_data.py e3diff sar2opt
hf download JeonghyeokDo/GeoSET --include "baselines/sar2opt/e3diff/*" --local-dir hf
mkdir -p $CKPT_ROOT/e3diff/sar2opt && cp hf/baselines/sar2opt/e3diff/gen.pth $CKPT_ROOT/e3diff/sar2opt/
bash baselines/e3diff/test.sh sar2opt        # writes $WORK_DIR/results/e3diff/sar2opt/<stem>.png
python scripts/evaluate.py --dataset sar2opt --pred-dir $WORK_DIR/results/e3diff/sar2opt

Licence

The E3Diff repository contains no licence file (its bundled SoftPool/ directory is MIT). These weights were trained by the GeoSET authors with that code; the GeoSET repository distributes only its changes to the code.