Pushing the Boundaries of Boundary Detection using Deep Learning

Abstract

In this work we show that adapting Deep Convolutional Neural Network training to the task of boundary detection can result in substantial improvements over the current state-of-the-art in boundary detection. Our contributions consist firstly in combining a careful design of the loss for boundary detection training, a multi-resolution architecture and training with external data to improve the detection accuracy of the current state of the art. When measured on the standard Berkeley Segmentation Dataset, we improve theoptimal dataset scale F-measure from 0.780 to 0.808 - while human performance is at 0.803. We further improve performance to 0.813 by combining deep learning with grouping, integrating the Normalized Cuts technique within a deep network. We also examine the potential of our boundary detector in conjunction with the task of semantic segmentation and demonstrate clear improvements over state-of-the-art systems. Our detector is fully integrated in the popular Caffe framework and processes a 320x420 image in less than a second.

The previous version reported large improvements w.r.t. the LPO region proposal baseline, which turned out to be due to a wrong computation for the baseline. The improvements are currently less important, and are omitted. We are sorry if the reported results caused any confusion. We have also integrated reviewer feedback regarding human performance on the BSD benchmark

References in corpus (6)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Depth Map Prediction from a Single Image using a Multi-Scale Deep Network
• Going Deeper with Convolutions
• Fully Convolutional Networks for Semantic Segmentation
• Semantic Image Segmentation via Deep Parsing Network
• Pixel-wise Deep Learning for Contour Detection

Cited by in corpus (20)

• Learning Deep Structured Multi-Scale Features using Attention-Gated CRFs for Contour Prediction
• Deep multi-task learning for a geographically-regularized semantic segmentation of aerial images
• Flood-Filling Networks
• Semantic Image Segmentation with Task-Specific Edge Detection Using CNNs and a Discriminatively Trained Domain Transform
• A Tropical Approach to Neural Networks with Piecewise Linear Activations
• LEGO: Learning Edge with Geometry all at Once by Watching Videos
• Pixel Difference Networks for Efficient Edge Detection
• SE2Net: Siamese Edge-Enhancement Network for Salient Object Detection
• Image-based localization using LSTMs for structured feature correlation
• Recent Advances in the Applications of Convolutional Neural Networks to Medical Image Contour Detection
• Three Birds One Stone: A General Architecture for Salient Object Segmentation, Edge Detection and Skeleton Extraction
• Segmenting Objects in Day and Night:Edge-Conditioned CNN for Thermal Image Semantic Segmentation
• BoundarySqueeze: Image Segmentation as Boundary Squeezing
• A Simple Pooling-Based Design for Real-Time Salient Object Detection
• Unmixing Convolutional Features for Crisp Edge Detection
• Better Image Segmentation by Exploiting Dense Semantic Predictions
• Piecewise Flat Embedding for Image Segmentation
• Hi-Fi: Hierarchical Feature Integration for Skeleton Detection
• Deep, Dense, and Low-Rank Gaussian Conditional Random Fields
• MoE-SPNet: A Mixture-of-Experts Scene Parsing Network