Datasets and trained diffusion models for "Diffusion Models for Interferometric Satellite Aperture Radar"

<p>A set of trained Probabilistic Diffusion Models (PDMs) and corresponding training datasets for the paper &quot;<a href="https://doi.org/10.48550/arXiv.2308.16847">Diffusion Models for Interferometric Satellite Aperture Radar</a>&quot;, by Tuel, Kerdreux et al. The code for this paper can be found at <a href="https://github.com/thomaskerdreux/PDM_SAR_InSAR_generation">this link</a>.</p> <p><strong>Training datasets</strong></p> <p>- &quot;InSAR_noise_32x32.zip&quot;: a dataset of 32x32 ground deformation scenes obtained from InSAR interferograms over New Mexico with the small baseline subset (SBAS) algorithm. Images were normalised to [0, 1].</p> <p>- &quot;insar_unwrapped_phase_normalised.zip&quot;: a dataset of 128x128 InSAR interferograms obtained from Sentinel-1 acquisitions over Nex Mexico. Images were normalised to [0, 1].</p> <p><strong>Trained Models</strong></p> <p>We provide 6 trained PDMs in separate .zip files. Each .zip file contains the model weights (in *.pt format) and the model metadata file (in *.json format).</p> <p>- &quot;mnist_32_cond_sigma_100.zip&quot;: a class-conditional model trained with 100 diffusion time steps on 32x32 MNIST images;</p> <p>- &quot;mnist_32_no_cond_sigma_100.zip&quot;: an unconditional model trained with 100 diffusion time steps on 32x32 MNIST images;</p> <p>- &quot;SAR_lowres_128_cond_sigma_2000.zip&quot;: a low-resolution (256 to 128) model trained with 2000 diffusion time steps on 128x128 TenGeoP-SARwv images;</p> <p>- &quot;SAR_superres_128_to_256_cond_sigma_2000.zip&quot;: a super-resolution (128 to 256) model trained with 2000 diffusion time steps on TenGeoP-SARwv images;</p> <p>- &quot;insar_phase_128_sigma_2000.zip&quot;: an unconditional model trained with 2000 diffusion time steps on 128x128 Sentinel-1 InSAR interferograms over New Mexico;</p> <p>- &quot;insar_noise_32_sigma_1000.zip&quot;: an unconditional model trained with 1000 diffusion time steps on 32x32 Sentinel-1 InSAR ground deformation scenes over New Mexico.</p>