Enhancing Edge Detection by Texture Handling Architecture and Noiseless Training Data

2024-07-29

AI Summary

Edge detection (ED) faces two key challenges: severe annotation noise in human-labeled ground truth severely limits model performance, and existing methods struggle to simultaneously achieve high accuracy and texture robustness under stringent error tolerances. To address these, we propose the Cascaded Skip Density Block (CSDB) architecture and a noise-agnostic training paradigm. CSDB enhances texture preservation through multi-scale density modeling, while our training strategy enables direct supervision using clean edge maps—rather than noisy manual annotations—for the first time, augmented by noise-suppressing data augmentation. Evaluated on standard benchmarks including BSDS500 and NYUDv2, our method achieves state-of-the-art performance, with significant improvements in average precision (AP). The source code is publicly released to ensure reproducibility and validate effectiveness.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Safety and Robustness

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Image edge detection (ED) is a fundamental task in computer vision. While convolution-based models have significantly advanced ED performance, achieving high precision under strict error tolerance constraints remains challenging. Furthermore, the reliance on noisy, human-annotated training data limits model performance, even when the inputs are edge maps themselves. In this paper, we address these challenges in two key aspects. First, we propose a novel ED model incorporating Cascaded Skipping Density Blocks (CSDB) to enhance precision and robustness. Our model achieves state-of-the-art (SOTA) performance across multiple datasets, with substantial improvements in average precision (AP), as demonstrated by extensive experiments. Second, we introduce a novel data augmentation strategy that enables the integration of noiseless annotations during training, improving model performance, particularly when processing edge maps directly. Our findings contribute to a more precise ED architecture and the first method for integrating noiseless training data into ED tasks, offering potential directions for improving ED models. Codes can be found on https://github.com/Hao-B-Shu/SDPED.

Problem

• Improving edge detection precision under strict error constraints
• Reducing reliance on noisy human-annotated training data
• Enhancing model performance with noiseless data augmentation

Innovation

• Cascaded Skipping Density Blocks enhance precision
• Noiseless training data improves model performance
• Novel data augmentation integrates clean annotations