Institutional Repository of Vilnius University
Development of a codenames pictures companion using machine learning models /
Abstract
dc:descriptionThis work investigates the development of an ML-based companion for the board game "Codenames: Pictures" using modern vision–language pre-training (VLP) models. The main objective of the work is to design and implement an autonomous system capable of generating and interpreting clues in a cooperative game setting. A developed game simulation system supporting both contrastive (CLIP) and generative (BLIP) model families. Experimental evaluation was conducted using black-and-white and colorful image collections to evaluate model performance. The results show that generative models produce more human-understandable clues, while contrastive models are more robust in clue interpretation. In addition, the study demonstrates that color information plays a critical role in the effectiveness of VLP-based Codenames companions. It was also found that model performance degrades when using black-and-white image collections: the generated clues become less accurate and more difficult to interpret, as models are not always able to correctly identify the depicted objects or actions in the images.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Nikulin, Arsenij,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD277691240&prefLang=en_US