On the unreasonable effectiveness of CNNs
This work provides an incremental analysis of CNN capabilities for a specific inverse problem, relevant to researchers in deep learning and image processing.
The paper tackled the problem of assessing the upper bounds of baseline CNNs for image-to-image tasks by applying a standard U-Net to XOR decryption from noisy data, achieving acceptable results.
Deep learning methods using convolutional neural networks (CNN) have been successfully applied to virtually all imaging problems, and particularly in image reconstruction tasks with ill-posed and complicated imaging models. In an attempt to put upper bounds on the capability of baseline CNNs for solving image-to-image problems we applied a widely used standard off-the-shelf network architecture (U-Net) to the "inverse problem" of XOR decryption from noisy data and show acceptable results.