Read the original document by opening this link in a new tab.
Table of Contents
1. Introduction
2. Holistically-Nested Edge Detection
2.1. Existing multi-scale and multi-level NN
2.2. Formulation
3. Network Architecture
Summary
The document discusses a new edge detection algorithm called Holistically-Nested Edge Detection (HED). It addresses important issues in edge detection such as holistic image training and prediction, multi-scale and multi-level feature learning. The algorithm leverages deep learning techniques like fully convolutional neural networks and deeply-supervised nets to automatically learn rich hierarchical representations for improved edge and object boundary detection. It significantly advances the state-of-the-art on datasets like BSD500 and NYU Depth. The paper explains the formulation of the approach, the training phase, and the network architecture of HED.