Tiling Generative Adversarial Networks for Large-Scale Texture Synthesis
Date:
Monday, September 2, 2019 14:30 - 15:30
Speaker:
Anna Frühstück (Visual Computing Center, KAUST)
Location:
Meeting room 2nd floor / Office Bldg West (I21.01.132)
Series:
Mathematics and CS Seminar
Host:
Bernd Bickel
Contact:
PETZ Caroline
High-quality textures are an important component in many applications ranging from architectural rendering to game design and often require the work of skilled artists. The synthesis of natural textures is therefore an important problem, but the complexity of texture synthesis increases exponentially with the target size of the output texture. Recent advances in the field of Generative Adversarial Networks allow for novel techniques in this field with unprecedented visual quality. TileGAN tackles the problem of texture synthesis in the setting where many input images are given and a large-scale output is required. We propose an algorithm to combine outputs of GANs to produce high-resolution texture maps with virtually no boundary artifacts. Our user interface enables artistic control and allows for an interactive guided synthesis.