| Summary: | 碩士 === 國立高雄科技大學 === 資訊工程系 === 107 === With the emergence of artificial intelligence, artificial neural networks have become a widely studied topic. One type of artificial neural network that is gaining popularity is the generative adversarial network (GAN) because it can demonstrate excellent performance through training and is applicable to a wide range of applications, such as avatar generation, change of style, high-resolution image generation, and image restoration. However, GAN training requires a large amount of data, and insufficient data can lead to mediocre results, such as blurred edges and poor distribution of colors. Therefore, this study aimed to achieve favorable training results using a small amount of data.
This study selected weapons as training data for applications in the gaming industry, which needs to generate a wide variety of weaponry according to game production requirements. Ability to generate weaponry compatible to players’classes and levels by simply changing specific parameters facilitates diversity in weapon models, rendering the game less dull and boring.
|
|---|
