Downloads · 30 days
0
geonusaf-stage2-mask2image-block-fold1-D
Pranilllllll/geonusaf-stage2-mask2image-block-fold1-D
geonusaf-stage2-mask2image-block-fold1-D is a machine learning model from Pranilllllll. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for diffusers.
Paints a Kathmandu Valley satellite image onto a 7-class land-use mask. Trained on the fold-1 TRAIN tiles only; validation tiles never enter a gradient.
Access
Public
Updated Aug 29, 2026
Repo size
29 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt2.2 GB · 96%
At a glance
- Library
- diffusers
- Access
- Public
- Created
- Aug 26, 2026
- Updated
- Aug 29, 2026
- SHA
- 384469fb
Model card
From the Hugging Face model README
GeoNUSAF Stage 2 - mask-conditioned satellite image generator (block fold 1, ablation D)
Paints a Kathmandu Valley satellite image onto a 7-class land-use mask. Trained on the fold-1 TRAIN tiles only; validation tiles never enter a gradient.
| field | value |
|---|---|
| base | stable-diffusion-v1-5/stable-diffusion-v1-5 (frozen) |
| control branch | lllyasviel/control_v11p_sd15_seg, warm-started, fine-tuned |
| condition | onehot, 7 channels, init zeros |
| LoRA | r16 alpha 16 on to_q/to_k/to_v/to_out.0 |
| lr | LoRA 0.0001, control 5e-05 |
| resolution | 512, GSD 0.586 m/px |
| sampling | DDIM 30 steps, guidance 3.5 |
| split | block fold 1, train_sha1 eb0aebdd9919 |
| seed | 42 |
| selection | layout (sel layout mIoU 0.4171) |
| steps | 7000 |
| layout mIoU (gen) | 0.5161 |
| layout mIoU (real ceiling) | 0.4643 |
| layout ratio | 1.112 |
| KID | 0.02629 +/- 0.00753 |
| FID | 122.44 |
Input contract
Single-channel class-index PNG, 512x512, values {0..5} with 255 (or 6) for ignore. Same format whether the mask is real or produced by Stage 1.
Output
genimage{i}.png + genmask{i}.png, paired, ready for segmentation training.
- Library
- diffusers
- Created
- Aug 26, 2026
- Updated
- Aug 29, 2026