W-Net: A Deep Model for Fully Unsupervised Image Segmentation

Abstract

While significant attention has been recently focused on designing supervised deep semantic segmentation algorithms for vision tasks, there are many domains in which sufficient supervised pixel-level labels are difficult to obtain. In this paper, we revisit the problem of purely unsupervised image segmentation and propose a novel deep architecture for this problem. We borrow recent ideas from supervised semantic segmentation methods, in particular by concatenating two fully convolutional networks together into an autoencoder--one for encoding and one for decoding. The encoding layer produces a k-way pixelwise prediction, and both the reconstruction error of the autoencoder as well as the normalized cut produced by the encoder are jointly minimized during training. When combined with suitable postprocessing involving conditional random field smoothing and hierarchical segmentation, our resulting algorithm achieves impressive results on the benchmark Berkeley Segmentation Data Set, outperforming a number of competing methods.

References in corpus (6)

• Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials
• ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation
• Learning Deconvolution Network for Semantic Segmentation
• Convolutional Feature Masking for Joint Object and Stuff Segmentation
• Xception: Deep Learning with Depthwise Separable Convolutions
• Simultaneous Detection and Segmentation

Cited by in corpus (32)

• U-Net and its variants for medical image segmentation: theory and applications
• Towards Goal-Oriented Semantic Signal Processing: Applications and Future Challenges
• Semi-Supervised Semantic Segmentation in Earth Observation: The MiniFrance Suite, Dataset Analysis and Multi-task Network Study
• Unsupervised Object Segmentation by Redrawing
• Pixel-Level Clustering Network for Unsupervised Image Segmentation
• HR-SAR-Net: A Deep Neural Network for Urban Scene Segmentation from High-Resolution SAR Data
• Real-time Streaming Video Denoising with Bidirectional Buffers
• Finding an Unsupervised Image Segmenter in Each of Your Deep Generative Models
• Hybrid Skip: A Biologically Inspired Skip Connection for the UNet Architecture
• SEIGAN: Towards Compositional Image Generation by Simultaneously Learning to Segment, Enhance, and Inpaint
• Emergence of Object Segmentation in Perturbed Generative Models
• A Semantic-based Medical Image Fusion Approach
• Unsupervised Medical Image Segmentation with Adversarial Networks: From Edge Diagrams to Segmentation Maps
• Identifying Mislabeled Images in Supervised Learning Utilizing Autoencoder
• BRULÈ: Barycenter-Regularized Unsupervised Landmark Extraction
• BiO-Net: Learning Recurrent Bi-directional Connections for Encoder-Decoder Architecture
• TricycleGAN: Unsupervised Image Synthesis and Segmentation Based on Shape Priors
• Deep Learning to See: Towards New Foundations of Computer Vision
• Prostate Segmentation using 2D Bridged U-net
• A Dynamically Weighted Loss Function for Unsupervised Image Segmentation
• Post Processing of image segmentation using Conditional Random Fields
• A Spatial Guided Self-supervised Clustering Network for Medical Image Segmentation
• CAggNet: Crossing Aggregation Network for Medical Image Segmentation
• Probabilistic Model of Visual Segmentation
• Generalizing electrocardiogram delineation -- Training convolutional neural networks with synthetic data augmentation
• PL-Net: Progressive Learning Network for Medical Image Segmentation
• Universal Perceptual Grouping
• DeepTEGINN: Deep Learning Based Tools to Extract Graphs from Images of Neural Networks
• Unsupervised Image Segmentation by Mutual Information Maximization and Adversarial Regularization
• Deep Superpixel Cut for Unsupervised Image Segmentation
• Metric-based Regularization and Temporal Ensemble for Multi-task Learning using Heterogeneous Unsupervised Tasks
• Learning Rich Representations For Structured Visual Prediction Tasks