Neural Architecture Search for Deep Image Prior
Kary Ho[1], Andrew Gilbert[1], Hailin Jin[2], John Collomosse[1,2] [1] Centre for Vision Speech and Signal Processing, University of Surrey [2] Creative Intelligence Lab, Adobe Research We present a neural architecture search (NAS) technique to enhance the performance of image de-noising, in-painting and super-resolution tasks under the recently proposed Deep Image Prior (DIP). We show that evolutionary search can automatically optimize the encoder-decoder (E-D) structure and meta-parameters of the DIP network, which serves as a content-specific prior to regularize these single image restoration tasks. Our binary representation encodes the design space for an asymmetric E-D network that typically converges to yield a content-specific DIP within 10-20 generations using a population size of $500$. The optimized architectures consistently improve upon the visual quality of classical DIP for a diverse range of photographic and artistic content.
Kary Ho[1], Andrew Gilbert[1], Hailin Jin[2], John Collomosse[1,2] [1] Centre for Vision Speech and Signal Processing, University of Surrey [2] Creative Intelligence Lab, Adobe Research We present a neural architecture search (NAS) technique to enhance the performance of image de-noising, in-painting and super-resolution tasks under the recently proposed Deep Image Prior (DIP). We show that evolutionary search can automatically optimize the encoder-decoder (E-D) structure and meta-parameters of the DIP network, which serves as a content-specific prior to regularize these single image restoration tasks. Our binary representation encodes the design space for an asymmetric E-D network that typically converges to yield a content-specific DIP within 10-20 generations using a population size of $500$. The optimized architectures consistently improve upon the visual quality of classical DIP for a diverse range of photographic and artistic content.