Culture Reflecting Artistic Fashion Design using Deep Learning and Assisting Custom Algorithm

Abstract

This thesis report presents deep learning model developed for designing culture reflecting fashion. Handle image generation with complex structure and multicolored was the problem of previous works and because of that it is difficult to implement existing deep learning framework for culture reflecting fashion design. To solve the failures of previous work we have developed Custom assisting algorithm for image feature extraction and transformation that enabled us to treat content and style information or feature of image independently. Culture dataset is developed. The way of representing culture on fashion design using sketch and cultural pattern is designed and implemented. Our model used sketch for fashion design like human and controlling sketch according is handled and presented. At the end results gained from our model is discussed and evaluation with baseline is presented. Generally, this is the first work by which culture reflecting on fashion is controlled and enabled. In our work sketch of an image is used to control representation of cultural heritages on fashion and to control local structure of an image and color patterns are used to control representation of other cultural elements like emotion and feelings. Using our custom assisting algorithm, a lot of problems could be solved since it is generic feature transformation and extraction algorithm.