Pixel Difference Unmixing Feature Networks for Edge Detection

The edge detection model based on deep learning significantly improves performance, but its generally high complexity requires a large pretrained Convolutional Neural Networks (CNNs) backbone, and hence memory computing power. To solve this problem, we carefully choose proper components for detection, introduce Multiscale Aware Fusion Module self-attention feature-unmixing loss function, propose lightweight network model, Pixel Difference Unmixing Feature (PDUF). backbone of proposed is designed to adopt skip long-short residual connection does not use pre-trained weights, straightforward hyper-parameter settings. Extensive experiments the BSDS, NYUD, Multi-cue datasets, found that has higher F-scores than current state-of-the-art models (those with fewer 1 million parameters) BSDS500 (ODS F-score 0.818), NYUDv2 depth datasets 0.767) Multi-Cue dataset 0.871(0.002)), similar performance compared some (with about 35 parameters).