Memory-Efficient Hierarchical Neural Architecture Search for Image Restoration

Recently, much attention has been spent on neural architecture search (NAS), aiming to outperform those manually-designed architectures high-level vision recognition tasks. Inspired by the success, here we attempt leverage NAS techniques automatically design efficient network for low-level image restoration In particular, propose a memory-efficient hierarchical (termed HiNAS) and apply it two such tasks: denoising super-resolution. HiNAS adopts gradient based strategies builds flexible space, including inner space outer space. They are in charge of designing cell deciding widths, respectively. For layer-wise sharing strategy, resulting more better performance. cell-sharing strategy save memory, considerably accelerate speed. The proposed method is both memory computation efficient. With single GTX1080Ti GPU, takes only about 1 h searching BSD-500 dataset 3.5 super-resolution structure DIV2K dataset. Experiments show that found have fewer parameters enjoy faster inference speed, while achieving highly competitive performance compared with state-of-the-art methods. Code available at: https://github.com/hkzhang91/HiNAS