Please use this identifier to cite or link to this item: http://repository.ukrida.ac.id//handle/123456789/2249
Title: SNPE-SRGAN: Lightweight Generative Adversarial Networks for Single-Image Super Resolution on Mobile Using SNPE Framework
Authors: Tampubolon, Hendrik
Setyoko, Aji
Purnamasari, Fanindia
Keywords: Engineering controlled terms
Deep learning; Graphics processing unit; Optical resolving power
Engineering uncontrolled terms
Adversarial networks; Computation power; High resolution; Learning models; Model performance; Neural-processing; State of the art; Super resolution
Engineering main heading
Cameras
Issue Date: 23-Jun-2021
Publisher: IOP Publishing Ltd
Abstract: An image resulting from a low-resolution (LR) camera on the mobile phone has lower quality than a high-resolution(HR) camera on a DSLR. Meanwhile, the HR camera is pricing if compared with the LR camera. How to achieve a single-image quality on LR camera likewise on HR camera becomes essential research in the past years. Addressing this issue can be done by upscaling a single LR image. Recently, the super-resolution generative adversarial network (SRGAN) model is one of the state-of-the-art super-resolution(SR)models employed on singleimage SR. However, implementing a deep learning model like SRGAN on a mobile device is challenging in computation power and resources. This study aims to develop a smaller and lower resources model while preserving single-image SR quality on mobile devices. To meet these objectives, we convert, quantize, and compress the SRGAN model on Snapdragon Neural Processing Engine (SNPE) as an example. We then validate the SRGAN on the DIV2K dataset on which improves the model performances. Besides, we conduct experiments on GPU, DSP environment. The experimental result confirmed that SNPE-SRGAN capable of achieves not only HR images' quality but also low latency by 0.06 second and smaller model by 1.7 Mb size running on DSP. Also, the SRGAN-DLC-Quantized running on GPU has a smaller size by 1.7 Mb and lower latency by 1.151 seconds compared with Non-quantized SRGAN-TensorFlow by 9.1 Mb and 1.608 seconds latency.
URI: http://repository.ukrida.ac.id//handle/123456789/2249
ISSN: 17426588
Appears in Collections:Laporan bkd



Items in UKRIDA Repository are protected by copyright, with all rights reserved, unless otherwise indicated.