Akademik Veri Yönetim Sistemi
8th International Congress on Human-Computer Interaction, Optimization and Robotic Applications, ICHORA 2026, Ankara, Türkiye, 21 - 23 Mayıs 2026, (Tam Metin Bildiri)
In this study, we addressed the issue of incorrect maxillomandibular structure formation in images generated by conditional generative models with dental masks, especially in cases with missing teeth. Using the TUFTS dataset in a single-class setting, conditions that support the maxillomandibular structure are incorporated into the process of generating panoramic images from dental masks through a model that produces maxillomandibular masks. MaxilloGAN is developed by utilizing blocks from the ConvNeXt and SPADE architectures, achieving $23 \times$ fewer FLOPs and $4 \times$ fewer parameters compared to U-Net. The model is capable of converting single-channel binary dental masks into single-channel binary maxillomandibular region masks. The ratio of the real maxillomandibular region to the dental region and the ratio of the synthetic maxillomandibular region to the dental image are found to be sufficiently close to each other. Similarly, the proportions of real and synthetic maxillomandibular regions are also observed to be highly similar. It is further verified that all generated regions form a single connected structure.