QCNet: query context network for salient object detection of automatic surface inspection

Abstract Building upon fully convolutional networks (FCNs), deep learning-based salient object detection (SOD) methods achieve gratifying performance in many vision tasks, including surface defect detection. However, most existing FCN-based still suffer from the coarse edge predictions. The state-of-the-art employ intricate feature aggregation techniques to refine boundaries, but they are often too computational cost deploy real application. This paper proposes a semantics guided paradigm for Guided atrous pyramid module is first applied on top segment complete semantics. Query context modules further used build relation maps between saliency and structural information top-down pathway. These two allow semantic features flow throughout decoder phase, yielding detail enriched Experimental results demonstrate that proposed method performs favorably against SOD benchmarks. In addition, this can detect at 27 FPS fashion without any post-processing, which has potential real-time