Texture Synthesis with Recurrent Variational Auto-Encoder
This work addresses texture synthesis for computer vision applications, but it appears incremental as it builds on existing methods with a new loss function.
The paper tackles texture synthesis by proposing a recurrent variational auto-encoder with a novel loss function called FLTBNK, which is rotational and partially color invariant, and it generates neighboring tiles to expand textures, evaluated on the Describable Textures Dataset with both quantitative and qualitative experiments.
We propose a recurrent variational auto-encoder for texture synthesis. A novel loss function, FLTBNK, is used for training the texture synthesizer. It is rotational and partially color invariant loss function. Unlike L2 loss, FLTBNK explicitly models the correlation of color intensity between pixels. Our texture synthesizer generates neighboring tiles to expand a sample texture and is evaluated using various texture patterns from Describable Textures Dataset (DTD). We perform both quantitative and qualitative experiments with various loss functions to evaluate the performance of our proposed loss function (FLTBNK) --- a mini-human subject study is used for the qualitative evaluation.