| 요약2 |
This paper aims to generate and construct visual information in the field of interior design through the utilization of intelligent technologies. Specifically, it focuses on providing visual information images of bathroom spaces tailored to the user's physical characteristics, considering that safety incidents frequently occur in residential bathrooms.
Generative models within the continuously advancing field of artificial intelligence have increasingly found applications in the realm of design. When conducting tests for generating bathroom images using generative artificial intelligence capable of rapidly providing diverse alternatives for interior space images, it was observed that the specific physical characteristics of the user, as input, were not adequately reflected in the generated images.
Additionally, errors were identified in the forms of safety equipment within the generated bathroom images.
Therefore, in this paper, we constructed a dataset that included self-generated bathroom space images reflecting physical characteristics, along with accompanying text files that described each image. Subsequently, we conducted additional training using this dataset. The additional training was performed using the LoRA(Low-Rank Adaptation) method and took approximately 20 minutes. As a result of this training, a model file of 144MB was generated.
Images generated using this model demonstrated a notable contrast to the limitations of the original model. They featured correctly positioned safety equipment, and a significant reduction in errors was observed in the generated images. Through this content, it is anticipated that future developments in the comprehensiveness of incorporating physical characteristics and the construction of relevant information and training data will enhance the scalability and utility of image generation AI in the field of interior space design. |