- 學位論文
以人工智慧輔助圖標創造
Leveraging Artificial Intelligence to Facilitate Icon Creation
摘要
圖標（icon）被廣泛的應用在橫幅、標誌牌、廣告牌、網頁，以及行動應用程式中。我們介紹了一個系統，用以幫助設計師創造圖標。創造圖標時，設計師負責繪畫輪廓，而我們的系統則依此創造出不同風格的結果，如：休閒的、狂野的，以及優雅的。為了達到這個目標，我們使用我們收集的圖標資料集來訓練雙條件對抗式生成網路（dual conditional generative adversarial network ）。其中一個條件要求產生的圖片和畫出的輪廓要擁有相似的結構；另一個條件則要求此圖片和參考的圖標要擁有相似的風格。於是，生成器（generator）使用兩個輸入來產生圖標。兩個判別器（discriminator）則分別判斷產生的圖片是否滿足結構以及風格的限制。訓練完成的網路可以創造出設計師要求的圖標，並很好的減輕了他們的工作量。評估方面，我們以數個目前最先進的技術，和我們的雙條件對抗式生成網路做比較。為了促進學術研究，我們將會公開原始程式碼、圖標資料集，以及訓練好的網路。
並列摘要
We present a system to help designers create icons that are widely used in banners, signboards, billboards, homepages, and mobile apps. To create an icon, designers are tasked with drawing contours, whereas our system creates colorized results in different semantic styles, such as casual, wild, and elegant. This goal is achieved by training a dual conditional generative adversarial network (GAN) on our collected icon dataset. One condition requires the generated image and the drawn contour to possess a similar contour, while the other anticipates the image and the reference icon to have a similar style. Accordingly, the generator takes two inputs to create an icon image, and two discriminators determine whether the image fulfills the contour and style conditions. The trained network is able to create icons demanded by designers and greatly reduce their workload. For the evaluation, we compared our dual conditional GAN to several state-of-the-arts techniques. Experiment results demonstrate that our network is the best. Finally, we will provide the source code, icon dataset, and trained network for public use.