Extensions to the AutoEncoder tutorial

[tennlee]

Things to work:

• add more plotting of prediction and target side-by-side
• try things to improve thing
  • additional epochs
  • additional layers
  • larger latent space
• plot some accuracy scores for images using different metrics
  • Kernel Inception Distance
  • Structural Similarity Index
    • https://scikit-image.org/docs/0.25.x/auto_examples/transform/plot_ssim.html
    • https://stackoverflow.com/questions/53956932/use-pytorch-ssim-loss-function-in-my-model
  • Learned Perceptual Image Patch Similarity (LPIPS)
  • spectral loss?
• Try structural similarity index based loss function (1 - the SSIM)

Useful links:

• some articles on CNNs with satellite imagery
  • https://ieeexplore.ieee.org/document/9883256
  • https://arxiv.org/html/2505.00040v1
• metrics
  • https://huggingface.co/blog/PrunaAI/objective-metrics-for-image-generation-assessment
  • https://arxiv.org/html/2505.07175v1

Originally posted by @stevehadd in #188 (comment)