<?xml version="1.0" encoding="UTF-8" ?>
<modsCollection xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns="http://www.loc.gov/mods/v3" xmlns:slims="http://slims.web.id" xsi:schemaLocation="http://www.loc.gov/mods/v3 http://www.loc.gov/standards/mods/v3/mods-3-3.xsd">
<mods version="3.3" ID="20317">
<titleInfo>
<title><![CDATA[DILATED CONVOLUTION PADA RESIDUAL-IN-RESIDUAL DENSE BLOCK DI ESRGAN UNTUK UPSCALE IMAGE LOW RESOLUTION]]></title>
</titleInfo>
<name type="Personal Name" authority="">
<namePart>0415068801 - Yusup Miftahuddin, S.Kom., MT</namePart>
<role><roleTerm type="text">Dosen Pembimbing 1</roleTerm></role>
</name>
<name type="Personal Name" authority="">
<namePart>Nicola Vito / 152020049</namePart>
<role><roleTerm type="text">Penulis</roleTerm></role>
</name>
<name type="Personal Name" authority="">
<namePart>120240101 - Diash Firdaus, S.T., M.T.</namePart>
<role><roleTerm type="text">Dosen Pembimbing 2</roleTerm></role>
</name>
<typeOfResource manuscript="yes" collection="yes"><![CDATA[mixed material]]></typeOfResource>
<genre authority="marcgt"><![CDATA[bibliography]]></genre>
<originInfo>
<place><placeTerm type="text"><![CDATA[Teknik Informatika]]></placeTerm></place>
<publisher><![CDATA[FTI]]></publisher>
<dateIssued><![CDATA[2025]]></dateIssued>
<issuance><![CDATA[monographic]]></issuance>
<edition><![CDATA[0]]></edition>
</originInfo>
<language>
<languageTerm type="code"><![CDATA[en]]></languageTerm>
<languageTerm type="text"><![CDATA[English]]></languageTerm>
</language>
<physicalDescription>
<form authority="gmd"><![CDATA[Text]]></form>
<extent><![CDATA[]]></extent>
</physicalDescription>
<note>Citra dengan resolusi rendah sering mengalami penurunan kualitas seperti kehilangan detail halus, kabur, dan distorsi warna. Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) merupakan salah satu metode super-resolusi berbasis deep learning yang efektif dalam meningkatkan kualitas gambar. Penelitian ini mengusulkan modifikasi arsitektur ESRGAN dengan menerapkan dilated convolution pada Residual-in-Residual Dense Block (RRDB) guna memperluas receptive field tanpa meningkatkan jumlah parameter secara signifikan. Penelitian dilakukan menggunakan dataset DIV2K untuk pelatihan serta Set5 untuk pengujian. Evaluasi dilakukan dengan metrik PSNR, SSIM, dan LPIPS. Hasil pengujian menunjukkan bahwa ESRGAN original memperoleh nilai PSNR sebesar 30.83 dB, SSIM sebesar 0.8168, dan LPIPS sebesar 0.1946, yang menandakan kualitas rekonstruksi citra yang lebih tajam, lebih mirip dengan citra asli secara struktural, serta memiliki persepsi visual yang lebih baik. Sementara itu, ESRGAN dengan dilated convolution hanya mencapai PSNR sebesar 30.12 dB, SSIM sebesar 0.8009, dan LPIPS sebesar 0.2171, yang mencerminkan penurunan performa pada ketiga metrik. Dengan demikian, dapat disimpulkan bahwa penerapan dilated convolution pada RRDB belum mampu melampaui performa ESRGAN original, meskipun secara visual hasilnya tetap kompetitif.

Low-resolution images often suffer from quality degradation such as loss of fine details, blurriness, and color distortion. The Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) is one of the most effective deep learning-based super-resolution methods for improving image quality. This study proposes a modification of the ESRGAN architecture by applying dilated convolution in the Residual-in-Residual Dense Block (RRDB) to expand the receptive field without significantly increasing the number of parameters. The experiments were conducted using the DIV2K dataset for training and the Set5 dataset for testing. Evaluation was carried out using PSNR, SSIM, and LPIPS metrics. The experimental results show that the original ESRGAN achieved a PSNR of 30.83 dB, an SSIM of 0.8168, and an LPIPS of 0.1946, indicating sharper image reconstruction, better structural similarity to the ground truth, and higher perceptual quality. Meanwhile, ESRGAN with dilated convolution achieved a PSNR of only 30.12 dB, an SSIM of 0.8009, and an LPIPS of 0.2171, reflecting a decline in performance across all three metrics. Therefore, it can be concluded that applying dilated convolution in RRDB did not outperform the original ESRGAN, although the visual results remain competitive.</note>
<subject authority=""><topic><![CDATA[GAN, citra digital]]></topic></subject>
<subject authority=""><topic><![CDATA[ESRGAN, super-resolution, dilated convolution]]></topic></subject>
<classification><![CDATA[]]></classification><identifier type="isbn"><![CDATA[20250908]]></identifier><location>
<physicalLocation><![CDATA[Setiadi Open Source ETD System]]></physicalLocation>
<shelfLocator><![CDATA[920IF/25]]></shelfLocator>
<holdingSimple>
<copyInformation>
<numerationAndChronology type="1"><![CDATA[920IF/25]]></numerationAndChronology>
<sublocation><![CDATA[]]></sublocation>
<shelfLocator><![CDATA[920IF/25]]></shelfLocator>
</copyInformation>
</holdingSimple>
</location>
<slims:digitals>
<slims:digital_item id="22332" url="" path="/152020049_920IF.pdf" mimetype="application/pdf"><![CDATA[DILATED CONVOLUTION PADA RESIDUAL-IN-RESIDUAL DENSE BLOCK DI ESRGAN UNTUK UPSCALE IMAGE LOW RESOLUTION]]></slims:digital_item>
</slims:digitals><recordInfo>
<recordIdentifier><![CDATA[20317]]></recordIdentifier>
<recordCreationDate encoding="w3cdtf"><![CDATA[2025-11-10 14:30:38]]></recordCreationDate>
<recordChangeDate encoding="w3cdtf"><![CDATA[2025-11-11 13:07:13]]></recordChangeDate>
<recordOrigin><![CDATA[machine generated]]></recordOrigin>
</recordInfo></mods></modsCollection>