MEMO: Human-like Crisp Edge Detection Using Masked Edge Prediction

TL;DR

MEMO introduces a training and inference strategy that produces human-like, crisp edge detection results without complex loss functions or architecture modifications, by leveraging confidence gradients and progressive prediction.

Contribution

The paper demonstrates that a carefully designed training and inference approach with cross-entropy loss can achieve crisp, human-like edges, eliminating the need for specialized loss functions or network changes.

Findings

MEMO produces visually appealing, crisp edges without post-processing.

It outperforms prior methods on crispness-aware evaluations.

A large synthetic dataset improves generalization.

Abstract

Learning-based edge detection models trained with cross-entropy loss often suffer from thick edge predictions, which deviate from the crisp, single-pixel annotations typically provided by humans. While previous approaches to achieving crisp edges have focused on designing specialized loss functions or modifying network architectures, we show that a carefully designed training and inference strategy alone is sufficient to achieve human-like edge quality. In this work, we introduce the Masked Edge Prediction MOdel (MEMO), which produces both accurate and crisp edges using only cross-entropy loss. We first construct a large-scale synthetic edge dataset to pre-train MEMO, enhancing its generalization ability. Subsequent fine-tuning on downstream datasets requires only a lightweight module comprising 1.2\% additional parameters. During training, MEMO learns to predict edges under varying ratios of input masking. A key insight guiding our inference is that thick edge predictions typically exhibit a confidence gradient: high in the center and lower toward the boundaries. Leveraging this, we propose a novel progressive prediction strategy that sequentially finalizes edge predictions in order of prediction confidence, resulting in thinner and more precise contours. Our method achieves visually appealing, post-processing-free, human-like edge maps and outperforms prior methods on crispness-aware evaluations.