Add GeoSET release files (139/211)

f49a31e verified 22 days ago

2.77 kB

# C-DiffSET · SAR2Opt

C-DiffSET (TCSVT'26) retrained by the GeoSET authors on the SAR2Opt split. These are the weights behind
the C-DiffSET row for SAR2Opt in Table 2 of the GeoSET paper
([arXiv:2609.37496](https://arxiv.org/abs/2609.37496)).

- Method: Do, Lee, Lee and Kim, "C-DiffSET: Leveraging latent diffusion for SAR-to-EO image translation with confidence-guided reliable object generation", IEEE Transactions on Circuits and Systems for Video Technology, 2026.
- Implementation: [KAIST-VICLab/C-DiffSET](https://github.com/KAIST-VICLab/C-DiffSET) at commit `6fc380215ae97c9d134e8f1434061b652339b45c`, on Stable Diffusion 2.1-base (`Manojb/stable-diffusion-2-1-base`, revision `0094d483a120f3f33dafbd187ea4aa60d10de75c`).
- Training and inference code: [`baselines/c-diffset`](https://github.com/KAIST-VICLab/GeoSET/tree/main/baselines/c-diffset) in the GeoSET repository.

## Files

| File | Bytes | SHA-256 | Role |
|---|---:|---|---|
| `diffusion_pytorch_model.safetensors` | 3,463,784,116 | `2dd757db2b64206f4bb0bae04764fcfdb6c74b72fca8423a395b770c2ba7835f` | UNet weights (diffusers `UNet2DConditionModel`, fp32); input = SAR latent and noisy EO latent (8 channels), output = noise (channels 0-3) and a spatial variance whose inverse is the confidence (channel 4) |
| `config.json` | 911 | `a387cad525c0765440e96bad8b2ece4c9a63712c4df04ed95a95b57ddbd9927c` | UNet configuration |

## Results (GeoSET paper, Table 2)

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

| FID↓ | DISTS↓ | KID↓ | DINO↑ | LPIPS↓ | SSIM↑ | PSNR↑ |
|---:|---:|---:|---:|---:|---:|---:|
| 78.1 | 0.214 | 0.0138 | 0.601 | 0.529 | 0.314 | 16.81 |

## 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/c-diffset/fetch.sh
python baselines/data/ldm_prepare.py sar2opt
hf download JeonghyeokDo/GeoSET "baselines/sar2opt/c-diffset/diffusion_pytorch_model.safetensors" "baselines/sar2opt/c-diffset/config.json" --local-dir hf
mkdir -p $CKPT_ROOT/c-diffset/sar2opt && cp hf/baselines/sar2opt/c-diffset/* $CKPT_ROOT/c-diffset/sar2opt/
bash baselines/c-diffset/test.sh sar2opt        # writes $WORK_DIR/results/c-diffset/sar2opt/<stem>.png
python scripts/evaluate.py --dataset sar2opt --pred-dir $WORK_DIR/results/c-diffset/sar2opt
```

## Licence

These weights are a derivative of Stable Diffusion 2.1-base and are released under the CreativeML Open RAIL++-M licence ([text](https://huggingface.co/stabilityai/stable-diffusion-2/blob/main/LICENSE-MODEL)); the use-based restrictions of its Attachment A apply to any use and redistribution. The C-DiffSET code is MIT (Copyright (c) 2026 KAIST VICLab).

	# C-DiffSET · SAR2Opt

	C-DiffSET (TCSVT'26) retrained by the GeoSET authors on the SAR2Opt split. These are the weights behind
	the C-DiffSET row for SAR2Opt in Table 2 of the GeoSET paper
	([arXiv:2609.37496](https://arxiv.org/abs/2609.37496)).

	- Method: Do, Lee, Lee and Kim, "C-DiffSET: Leveraging latent diffusion for SAR-to-EO image translation with confidence-guided reliable object generation", IEEE Transactions on Circuits and Systems for Video Technology, 2026.
	- Implementation: [KAIST-VICLab/C-DiffSET](https://github.com/KAIST-VICLab/C-DiffSET) at commit `6fc380215ae97c9d134e8f1434061b652339b45c`, on Stable Diffusion 2.1-base (`Manojb/stable-diffusion-2-1-base`, revision `0094d483a120f3f33dafbd187ea4aa60d10de75c`).
	- Training and inference code: [`baselines/c-diffset`](https://github.com/KAIST-VICLab/GeoSET/tree/main/baselines/c-diffset) in the GeoSET repository.

	## Files

	\| File \| Bytes \| SHA-256 \| Role \|
	\|---\|---:\|---\|---\|
	\| `diffusion_pytorch_model.safetensors` \| 3,463,784,116 \| `2dd757db2b64206f4bb0bae04764fcfdb6c74b72fca8423a395b770c2ba7835f` \| UNet weights (diffusers `UNet2DConditionModel`, fp32); input = SAR latent and noisy EO latent (8 channels), output = noise (channels 0-3) and a spatial variance whose inverse is the confidence (channel 4) \|
	\| `config.json` \| 911 \| `a387cad525c0765440e96bad8b2ece4c9a63712c4df04ed95a95b57ddbd9927c` \| UNet configuration \|

	## Results (GeoSET paper, Table 2)

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

	\| FID↓ \| DISTS↓ \| KID↓ \| DINO↑ \| LPIPS↓ \| SSIM↑ \| PSNR↑ \|
	\|---:\|---:\|---:\|---:\|---:\|---:\|---:\|
	\| 78.1 \| 0.214 \| 0.0138 \| 0.601 \| 0.529 \| 0.314 \| 16.81 \|

	## 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/c-diffset/fetch.sh
	python baselines/data/ldm_prepare.py sar2opt
	hf download JeonghyeokDo/GeoSET "baselines/sar2opt/c-diffset/diffusion_pytorch_model.safetensors" "baselines/sar2opt/c-diffset/config.json" --local-dir hf
	mkdir -p $CKPT_ROOT/c-diffset/sar2opt && cp hf/baselines/sar2opt/c-diffset/* $CKPT_ROOT/c-diffset/sar2opt/
	bash baselines/c-diffset/test.sh sar2opt # writes $WORK_DIR/results/c-diffset/sar2opt/<stem>.png
	python scripts/evaluate.py --dataset sar2opt --pred-dir $WORK_DIR/results/c-diffset/sar2opt
	```

	## Licence

	These weights are a derivative of Stable Diffusion 2.1-base and are released under the CreativeML Open RAIL++-M licence ([text](https://huggingface.co/stabilityai/stable-diffusion-2/blob/main/LICENSE-MODEL)); the use-based restrictions of its Attachment A apply to any use and redistribution. The C-DiffSET code is MIT (Copyright (c) 2026 KAIST VICLab).