Fast, Accurate and Lightweight Super-Resolution with Neural Architecture Search

Abstract

Deep convolution neural networks demonstrate impressive results in the super-resolution domain. A series of studies concentrate on improving peak signal noise ratio (PSNR) by using much deeper layers, which are not friendly to constrained resources. Pursuing a trade-off between the restoration capacity and the simplicity of models is still non-trivial. Recent contributions are struggling to manually maximize this balance, while our work achieves the same goal automatically with neural architecture search. Specifically, we handle superresolution with a multi-objective approach. We also propose an elastic search tactic at both micro and macro level, based on a hybrid controller that profits from evolutionary computation and reinforcement learning. Quantitative experiments help us to draw a conclusion that our generated models dominate most of the state-of-the-art methods with respect to the individual FLOPS.

DARTS-: Robustly Stepping out of Performance Collapse Without Indicators

“…Neural architecture search and DARTS variants. Over the years, researchers have sought to automatically discover neural architectures for various deep learning tasks to relieve humans from the tedious effort, ranging from image classification, objection detection, image segmentation to image super-resolution Ghiasi et al, 2019;Chu et al, 2020a) etc. Among many proposed approaches, Differentiable Architecture Search (DARTS) (Liu et al, 2019b) features weight-sharing and resolves the searching problem via gradient descent, which is very efficient and easy to generalize.…”

NAS-DIP: Learning Deep Image Prior with Neural Architecture Search

“…NAS applications and search space. NAS algorithms have been successfully applied to various tasks, including image classification [53,61,63,64,89,90], semantic segmentation [14,51,58], object detection [24], image restoration [17,72], and image generation [26]. In the context of object detection, NAS-FPN [24] develops a search space that allows the model to learn pyramidal representations by merging cross-scale features for improved detection of multiple objects with different scales and locations.…”

Searching for Controllable Image Restoration Networks

“…For deblurring, a method using spatially variant deconvolution is proposed in [26] to achieve accurate performance with its efficient backbone network. Meanwhile, FALSR [3], ESRN [19], and FGNAS [10] adopt neural architecture search (NAS) algorithms for efficient superresolution networks. Path-Restore [25] and AdaDSR [15] save computation costs via adaptive inference for general image restoration and super-resolution, respectively.…”