Issues with Output and Inference Time for SAR2SAR Denoising

[shailzasharma]

Hi,

Thank you for the great work on SAR2SAR.

I am trying to despeckle a SAR image using the SAR2SAR method. However, I'm encountering issues with the output images appearing blank. Here is the script I'm using with the attached image

path_to_tiff_image = '/content/20160901.slc.tif'
tiff_image_array = read_tiff_with_rasterio(path_to_tiff_image)
patch_size = 256
stride_size = 32
denoiser = Sar2SarDenoiser()
denoised_images = denoiser.denoise_image(tiff_image_array, patch_size, stride_size)

I'm experiencing two main issues:

The output images are appeared to be blank.
The inference time is approximately 1.5 hours when using the CPU for the attached image. Is this normal, or is there a mistake on our end?
Could you provide guidance on how to correctly use the arguments? Additionally, is the code compatible with GPU acceleration to potentially reduce inference time?

Your assistance would be greatly appreciated.

Best regards,
Shailza Sharma

[Tnecniv-Teikram]

Did you try converting your .tif file into .cos or .npy format? I think those are the only formats the model accepts.

This is what is stated in the readme:
To despeckle SAR images using MERLIN, images need to be in .cos or .npy format.