Multimodal Transfer: A Hierarchical Deep Convolutional Neural Network for Fast Artistic Style Transfer

Transferring artistic styles onto everyday photographs has become an extremely popular task in both academia and industry since the salient work by Gatys et al. More recently, several feed-forward networks were proposed, leading to significant speed up of the stylization process to nearly real-time by replacing the original online iterative optimization procedure with offline training. However, when those stylization networks are applied directly to high-resolution images, the style of localized regions often appears less similar to the desired artistic style, because the transfer process fails to capture small, intricate textures and maintain correct texture scales of the artworks. Here we propose a multimodal convolutional neural network that takes into consideration faithful representations of both color and luminance channels, and performs stylization hierarchically with multiple losses of increasing scales. Compared to the state-of-the-art networks, our network can also perform style transfer in nearly real time by performing much more sophisticated training offline. Furthermore, by properly handling style and texture cues at multiple scales using several modalities, we can transfer not just large-scale, obvious style cues but also subtle, exquisite ones. That is, our scheme can generate results that are visually pleasing and more similar to multiple desired artistic styles with color and texture cues at multiple scales.