Convolutional Oriented Boundaries: From Image Segmentation to High-Level Tasks (2016-08-09T00:00:00.000000Z)

TL;DR

Convolutional Oriented Boundaries gives a significant leap in performance over the state-of-the-art, and generalizes very well to unseen categories and datasets, and learning to estimate not only contour strength but also orientation provides more accurate results.

Abstract

We present Convolutional Oriented Boundaries (COB), which produces multiscale oriented contours and region hierarchies starting from generic image classification Convolutional Neural Networks (CNNs). COB is computationally efficient, because it requires a single CNN forward pass for multi-scale contour detection and it uses a novel sparse boundary representation for hierarchical segmentation; it gives a significant leap in performance over the state-of-the-art, and it generalizes very well to unseen categories and datasets. Particularly, we show that learning to estimate not only contour strength but also orientation provides more accurate results. We perform extensive experiments for low-level applications on BSDS, PASCAL Context, PASCAL Segmentation, and NYUD to evaluate boundary detection performance, showing that COB provides state-of-the-art contours and region hierarchies in all datasets. We also evaluate COB on high-level tasks when coupled with multiple pipelines for object proposals, semantic contours, semantic segmentation, and object detection on MS-COCO, SBD, and PASCAL; showing that COB also improves the results for all tasks.

Authors

J. Pont-Tuset

11 papers

Kevis-Kokitsi Maninis

4 papers

P. Arbeláez

4 papers

TL;DR

Abstract

Authors

References83 items

Pyramid Scene Parsing Network

UberNet: Training a Universal Convolutional Neural Network for Low-, Mid-, and High-Level Vision Using Diverse Datasets and Limited Memory

Deep Retinal Image Understanding

Supervised Evaluation of Image Segmentation and Object Proposal Techniques

R-FCN: Object Detection via Region-based Fully Convolutional Networks

Learning to Refine Object Segments

Object Contour Detection with a Fully Convolutional Encoder-Decoder Network

Deep Residual Learning for Image Recognition

SSD: Single Shot MultiBox Detector

Boosting Object Proposals: From Pascal to COCO

The Middle Child Problem: Revisiting Parametric Min-Cut and Seeds for Object Proposals

Weakly Supervised Object Boundaries

Multi-Scale Context Aggregation by Dilated Convolutions

Pushing the Boundaries of Boundary Detection using Deep Learning

Unsupervised Learning of Edges

Semantic Segmentation with Boundary Neural Fields

Learning to Segment Object Candidates

You Only Look Once: Unified, Real-Time Object Detection

Learning to propose objects

DeepContour: A deep convolutional feature learned by positive-sharing loss for contour detection

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

Fast R-CNN

Situational object boundary detection

Holistically-Nested Edge Detection

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and Its Applications to High-Level Vision

Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation

What Makes for Effective Detection Proposals?

Oriented edge forests for boundary detection

DeepEdge: A multi-scale bifurcated deep network for top-down contour detection

Deeply-Supervised Nets

Going deeper with convolutions

Edge Boxes: Locating Object Proposals from Edges

Geodesic Object Proposals

Crisp Boundary Detection Using Pointwise Mutual Information

Very Deep Convolutional Networks for Large-Scale Image Recognition

ImageNet Large Scale Visual Recognition Challenge

Learning Rich Features from RGB-D Images for Object Detection and Segmentation

N4-Fields: Neural Network Nearest Neighbor Fields for Image Transforms

Generating Object Segmentation Proposals Using Global and Local Search

The Role of Context for Object Detection and Semantic Segmentation in the Wild

RIGOR: Reusing Inference in Graph Cuts for Generating Object Regions

Fast Edge Detection Using Structured Forests

Caffe: Convolutional Architecture for Fast Feature Embedding

BING: Binarized normed gradients for objectness estimation at 300fps

Microsoft COCO: Common Objects in Context

Structured Forests for Fast Edge Detection

Prime Object Proposals with Randomized Prim's Algorithm

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

Perceptual Organization and Recognition of Indoor Scenes from RGB-D Images

Image Segmentation by Cascaded Region Agglomeration

Sketch Tokens: A Learned Mid-level Representation for Contour and Object Detection

Selective Search for Object Recognition

ImageNet classification with deep convolutional neural networks

Discriminatively Trained Sparse Code Gradients for Contour Detection

Indoor Segmentation and Support Inference from RGBD Images

CPMC: Automatic Object Segmentation Using Constrained Parametric Min-Cuts

Semantic contours from inverse detectors

Measuring the Objectness of Image Windows

Contour Detection and Hierarchical Image Segmentation

Boundary Detection Using F-Measure-, Filter- and Feature- (F3) Boost

Multi-scale Improves Boundary Detection in Natural Images

Supervised Learning of Edges and Object Boundaries

Efficient Graph-Based Image Segmentation

Learning to detect natural image boundaries using local brightness, color, and texture cues

Mean Shift: A Robust Approach Toward Feature Space Analysis

A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics

Normalized cuts and image segmentation

Geodesic Saliency of Watershed Contours and Hierarchical Segmentation

A Computational Approach to Edge Detection

On the accuracy of the Sobel edge detector

Theory of edge detection

Machine Perception of Three-Dimensional Solids

Convolutional Oriented Boundaries

Segmenting natural images with the least effort as humans

He received the degree in Mathematics in 2008, the degree in Electrical Engineering in 2008, the M.Sc

Luc Van Gool's Computer Vision Lab