Enhance Convolutional Neural Networks with Noise Incentive Block
This work tackles the problem of generating spatially-variant outputs from flat inputs for CNN models, which is a previously underexplored issue in image generation and translation tasks.
This paper addresses the 'flatness degradation' in CNNs, where spatially shared convolution kernels lead to less vivid results when fed with flat inputs. The authors propose the Noise Incentive Block (NIB), a model-agnostic plug-in that perturbs input data with a noise map and reassembles them in the feature domain, enabling CNNs to generate visually better results with richer details in tasks like semantic image synthesis and deep neural dithering.
As a generic modeling tool, Convolutional Neural Networks (CNNs) have been widely employed in image generation and translation tasks. However, when fed with a flat input, current CNN models may fail to generate vivid results due to the spatially shared convolution kernels. We call it the flatness degradation of CNNs. Unfortunately, such degradation is the greatest obstacles to generate a spatially-variant output from a flat input, which has been barely discussed in the previous literature. To tackle this problem, we propose a model agnostic solution, i.e. Noise Incentive Block (NIB), which serves as a generic plug-in for any CNN generation model. The key idea is to break the flat input condition while keeping the intactness of the original information. Specifically, the NIB perturbs the input data symmetrically with a noise map and reassembles them in the feature domain as driven by the objective function. Extensive experiments show that existing CNN models equipped with NIB survive from the flatness degradation and are able to generate visually better results with richer details in some specific image generation tasks given flat inputs, e.g. semantic image synthesis, data-hidden image generation, and deep neural dithering.