Enhancing Photogrammetric Digital Surface Models with Pretrained Diffusion Models and Multimodal Conditioning
Researchers refine satellite photogrammetric surface models with a modified Stable Diffusion 3 model conditioned on imagery.
Large-scale digital surface models from stereo satellite photogrammetry are often noisy and incomplete, while aerial LiDAR is more accurate but costly. The authors condition a modified Stable Diffusion 3 model, with a pruned text stream and patch-wise normalization, on photogrammetric DSMs and Pléiades imagery to refine co-registered elevation maps. In French cities, multimodal conditioning cut dense-urban RMSE from 6.00 to 3.45 m in-context and from 4.16 to 2.77 m in held-out Bordeaux.
- Modified Stable Diffusion 3 uses a pruned text stream and patch-wise normalization
- Training transfers from natural images onto LiDAR elevation maps
- Dense-urban RMSE falls from 6.00 m to 3.45 m in studied cities
- Held-out Bordeaux RMSE improves from 4.16 m to 2.77 m
- Tags
- diffusion-modelsstable-diffusion-3photogrammetrydigital-surface-modellidarremote-sensingpleiades
- Products
- Stable Diffusion 3Pléiades
- Countries
- France
- AI models
- Stable Diffusion 3
From the original article
1 min read at huggingface.co
Large-scale Digital Surface Models (DSMs) can be produced cost-effectively from satellite images via stereo-photogrammetry. However, the resulting 3D maps are often contaminated by noise, outliers, and voids. On the other hand, aerial LiDAR provides high-accuracy elevation measurements at a substantially higher cost.
By Antoine Lorentz, Stéphane May, Valentine Bellet · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.