The paper introduces a new edge detection algorithm called Holistically-Nested Edge Detection (HED), which addresses two key issues in edge detection: holistic image training and prediction, and multi-scale and multi-level feature learning. HED uses a deep learning model that leverages fully convolutional neural networks and deeply-supervised nets to perform image-to-image prediction. The method automatically learns rich hierarchical representations, which are crucial for resolving the ambiguity in edge and object boundary detection. HED significantly improves performance on the BSD500 and NYU Depth datasets, achieving an ODS F-score of 0.782 and 0.746, respectively, while being orders of magnitude faster than some recent CNN-based edge detection algorithms. The paper also discusses the architecture of HED, its implementation details, and experimental results, highlighting the effectiveness of deep supervision and multi-scale learning in achieving state-of-the-art performance.The paper introduces a new edge detection algorithm called Holistically-Nested Edge Detection (HED), which addresses two key issues in edge detection: holistic image training and prediction, and multi-scale and multi-level feature learning. HED uses a deep learning model that leverages fully convolutional neural networks and deeply-supervised nets to perform image-to-image prediction. The method automatically learns rich hierarchical representations, which are crucial for resolving the ambiguity in edge and object boundary detection. HED significantly improves performance on the BSD500 and NYU Depth datasets, achieving an ODS F-score of 0.782 and 0.746, respectively, while being orders of magnitude faster than some recent CNN-based edge detection algorithms. The paper also discusses the architecture of HED, its implementation details, and experimental results, highlighting the effectiveness of deep supervision and multi-scale learning in achieving state-of-the-art performance.