Institutional Repository of Vilnius University
Vaizdų dublikatų aptikimas konvoliuciniais neuroniniais tinklais /
Abstract
dc:descriptionIn this bachelor’s thesis, methods for detecting image duplicates based on convolutional neural networks are examined. Given the rapidly increasing amount of digital images on the internet, along with associated use cases and potential legal issues, existing duplicate detection methods (Multigrain, HOW+ASMK, and GIST) are analyzed, evaluated, and compared to determine which method most accurately and efficiently identifies image duplicates. This work also focuses on enhancing and adapting the CLIP model, which utilizes the ResNet50 neural network architecture, for the duplicate detection task, aiming to assess whether a modern image processing model can be effectively adapted in such a diverse environment. During the training and evaluation of the methods, the DISC21 dataset was used, and it was determined that the Multigrain method proved to be the most effective for this type of task, achieving the best results, while the traditional GIST method performed the worst. Upon finetuning and adapting the CLIP model for the image duplicate recognition task, it was found that training with contrastive or circle loss, combined with AdamW optimization, yielded better detection accuracy for the CLIP model compared to training done with a triplet loss function and RMSprop optimization.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Buračevskis, Robertas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD239030004&prefLang=en_US