AI AND ARTS

ANISOTROPIC SEMANTIC PRESERVING MULTIPLE ARTISTS STYLE TRANSFER

we propose an Anisotropic Semantic-Preserving Multi-Artist Style Transfer framework. On the one hand, we develop a multicondition single-generator structure which frst performs multi-artist style transfer. On the other hand, we introduce an Anisotropic Semantic Preserving Module (ASP Module), which endows the network with the ability of adjustable semantic consistency among various styles.

COMPOSITION-COMPATIBLE STILL-LIFE ARTWORK GENERATION WITH MASKGAN

we propose a novel framework to generate still-life artworks with realistic composition, through joint learning for objects’ appearance and their spatial relationships. The proposed framework features three building blocks. it takes the description text of the painting as input, converts it to scene graph, and obtains the spatial dependency information by graph convolution.