Image to Image Transfer Makes Malpositioned Teeth Orderly

Sanbi Luo

Abstract: Deep learning’s continuous development has spawned many applications and neural networks that tackle challenging tasks in different fields. The fusion of these applications and networks could solve more complex problems in the real world. Taking image generation as an example, combining a pure image generator by CNN and a single text processor by RNN can form a more complex and capable fusion network for text to image generation. Generally, a basic successful deep learning application requires at least three elements: a suitable application scenario, an appropriate dataset, and a proper model. Under this principle, in this paper, we introduce a new task for virtual orthodontics, a new image-to-image transfer task from malpositioned-teeth-image to neat-teeth-image. We call it orthodontics transfer. To make up for the lack of datasets, we constructed a paired dataset about orthodontics before and after surgery from real medical cases. Simultaneously, a new orthodontics transfer network with a teeth-code transfer as a bridge is proposed. The experimental results show that our proposed method is effective, which can realize the orthodontic effect on teeth photos by image-to-image transfer.

External IDs:dblp:conf/isbra/Luo21