Learning to predict crisp boundaries

By Ruoxi Deng, Chunhua Shen, Shengjun Liu, Huibing Wang, Xinru Liu. 2018

291 citations recorded by the source corpus.

Recent methods for boundary or edge detection built on Deep Convolutional Neural Networks (CNNs) typically suffer from the issue of predicted edges being thick and need post-processing to obtain crisp boundaries. Highly imbalanced categories of boundary versus background in training data is one of main reasons for the above problem. In this work, the aim is to make CNNs produce sharp boundaries without post-processing. We introduce a novel loss for boundary detection, which is very effective for classifying imbalanced data and allows CNNs to produce crisp boundaries. Moreover, we propose an end-to-end network which adopts the bottom-up/top-down architecture to tackle the task. The proposed network effectively leverages hierarchical features and produces pixel-accurate boundary mask, which is critical to reconstruct the edge map. Our experiments illustrate that directly making crisp prediction not only promotes the visual results of CNNs, but also achieves better results against the state-of-the-art on the BSDS500 dataset (ODS F-score of .815) and the NYU Depth dataset (ODS F-score of .762).

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

EDTER: Edge Detection with Transformer (2022, methodology): “[49], gPb+NG [19], SE [15], SE+NG+ [20], OEF [21], SemiContour [69], HED [65], RCF [36], AMH-Net [66], LPCB [12], BDCN [22], and PiDiNet [53].”
Rethinking BiSeNet For Real-time Semantic Segmentation (2021, methodology): “Because weighted cross-entropy always leads to coarse results, following [7], we adopt binary cross-entropy and dice loss to jointly optimize the detail learning.”
PIDNet: A Real-time Semantic Segmentation Network Inspired by PID Controllers (2022, background): “Instead of dice loss [13], weighted binary cross entropy loss l1 is adopted to deal with the imbalanced problem of boundary detection since coarse boundary is preferred to highlight the boundary region and enhance the features for small objects.”
Deepstrip: High-Resolution Boundary Refinement (2020, background): “Multiple attempts have been made to improve boundary quality through extracting better features [41, 43, 26, 2, 14].”
Mask3D: Mask Transformer for 3D Semantic Instance Segmentation (2022, background): “where LBCE is the binary cross-entropy loss (over the foreground and background of that mask) and Ldice is the Dice loss [10].”
Bi-Directional Cascade Network for Perceptual Edge Detection (2019, unknown): “LPCB [9] RGB .”
RINDNet: Edge Detection for Discontinuity in Reflectance, Illumination, Normal and Depth (2021, background): “CNN-based edge detectors [3, 4, 9, 10, 17, 20, 24, 26, 31, 36, 45] do not rely on hand-crafted features and achieve better performance.”
Deep Interactive Thin Object Selection (2021, methodology): “In this work, we adopt a combination of balanced binary cross-entropy loss [48] and Dice loss [7] for training the edge stream.”
Deep Structural Contour Detection (2020, methodology): “In recent years, benefited from the rapid development of the deep learning technologies [12, 13, 20], the performance of contour detection has been significantly improved [8, 46].”