Superpixel Segmentation via Convolutional Neural Networks with Regularized Information Maximization

TL;DR

This paper introduces an unsupervised superpixel segmentation method using a CNN optimized during inference, which adapts to images and controls superpixel properties without requiring labels.

Contribution

It presents a novel inference-time CNN optimization approach for superpixel segmentation that is unsupervised and adaptable to individual images.

Findings

Effective superpixel segmentation on BSDS500 and SBD datasets.

Advantages include image prior utilization, adaptive superpixel count, and property control.

Qualitative and quantitative validation of method's effectiveness.

Abstract

We propose an unsupervised superpixel segmentation method by optimizing a randomly-initialized convolutional neural network (CNN) in inference time. Our method generates superpixels via CNN from a single image without any labels by minimizing a proposed objective function for superpixel segmentation in inference time. There are three advantages to our method compared with many of existing methods: (i) leverages an image prior of CNN for superpixel segmentation, (ii) adaptively changes the number of superpixels according to the given images, and (iii) controls the property of superpixels by adding an auxiliary cost to the objective function. We verify the advantages of our method quantitatively and qualitatively on BSDS500 and SBD datasets.