Superpixel Segmentation with Fully Convolutional Networks
This addresses the challenge of using superpixels in deep learning for computer vision, offering a faster and more accurate method for tasks like dense prediction, though it is incremental in adapting existing network architectures.
The paper tackles the problem of efficiently incorporating superpixels into deep neural networks by proposing a fully convolutional network that predicts superpixels on a regular grid, achieving state-of-the-art performance at about 50fps. It further applies this to improve disparity estimation accuracy in stereo matching tasks.
In computer vision, superpixels have been widely used as an effective way to reduce the number of image primitives for subsequent processing. But only a few attempts have been made to incorporate them into deep neural networks. One main reason is that the standard convolution operation is defined on regular grids and becomes inefficient when applied to superpixels. Inspired by an initialization strategy commonly adopted by traditional superpixel algorithms, we present a novel method that employs a simple fully convolutional network to predict superpixels on a regular image grid. Experimental results on benchmark datasets show that our method achieves state-of-the-art superpixel segmentation performance while running at about 50fps. Based on the predicted superpixels, we further develop a downsampling/upsampling scheme for deep networks with the goal of generating high-resolution outputs for dense prediction tasks. Specifically, we modify a popular network architecture for stereo matching to simultaneously predict superpixels and disparities. We show that improved disparity estimation accuracy can be obtained on public datasets.