Self-supervised Learning untuk Reidentifikasi Orang Menggunakan SimCLR

Zhafiri, Radifan Naufal (2022) Self-supervised Learning untuk Reidentifikasi Orang Menggunakan SimCLR. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 07211840000018-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Sistem keamanan dengan menggunakan kamera CCTV banyak digunakan di ruang publik untuk pemantauan dan menganalisa tindakan kriminal. Salah satu permasalahan dari sistem keamanan ini adalah proses analisa hasil rekaman CCTV membutuhkan waktu lama dan rentan akan kesalahan manusia. Untuk membuat proses analisa hasil rekaman lebih akurat dan efisien dapat menggunakan teknologi reidentifikasi orang. Reidentifikasi orang adalah proses pencarian seseorang dari kumpulan data yang ditangkap oleh beberapa kamera berbeda dengan menggunakan teknik visi komputer dan Deep Learning. Tugas akhir ini dikembangkan menggunakan metode Self-supervised learning . Self-supervised memanfaatkan data tanpa label pada pembuatan model citra untuk mempelajari fitur dari data dan melakukan transfer hasil tersebut ke model downstream task. Hasil mAP yang didapat pada dataset Market-1501 ketika menggunakan self-supervised learning adalah 67.52% pada ResNet-50 dan 75.46% pada OSNet, sedangkan pada dataset DukeMTMC-Reid didapat 52.62% pada ResNet-50 dan 65.82% pada OSNet. Hasil yang didapat mengalami peningkatan jika dibandingkan model scratch pada Market -1501 58.2% & 64.32% dan pada DukeMTMC-Reid 47.6% & 55.3% pada backbone ResNet-50 dan OSNet.
==============================================================================================================================
Security systems using CCTV cameras are widely used in public spaces for monitoring and analyzing criminal acts. One of the problems with this security system is that the process of analyzing CCTV footage takes a long time and is prone to human error. To make the analyzing process more accurate and efficient, we can use person re-identification technology. Person re-identification is the process of searching for a person from a collection of data captured by several different cameras using computer vision and deep learning techniques. This final project was developed using the self-supervised learning method. Self-supervised utilizes unlabeled data in image modeling to learn features of the data and transfers the results to the downstream task model. The mAP results obtained in the Market-1501 dataset when using self-supervised learning were 67.52% on ResNet-50 and 75.46% on OSNet, while the DukeMTMC-Reid dataset obtained 52.62% on ResNet-50 and 65.82% on OSNet. The results obtained have increased when compared to the scratch model on Market -1501 58.2% & 64.32% and on DukeMTMC-Reid 47.6% & 55.3% on ResNet-50 and OSNet backbones.

Item Type:	Thesis (Other)
Additional Information:	RSKom 006.42 Zha s-1 2022
Uncontrolled Keywords:	Self-supervised Learning. Person Re-identification.
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science. EDP
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Mr. Marsudiyana -
Date Deposited:	17 Jun 2026 01:07
Last Modified:	17 Jun 2026 01:07
URI:	http://repository.its.ac.id/id/eprint/133828

Actions (login required)

View Item