Revisiting Image Pyramid Structure for High Resolution Salient Object Detection

Salient object detection (SOD) has been in the spotlight recently, yet hasbeen studied less for high-resolution (HR) images. Unfortunately, HR images andtheir pixel-level annotations are certainly more labor-intensive andtime-consuming compared to low-resolution (LR) images and annotations.Therefore, we propose an image pyramid-based SOD framework, Inverse SaliencyPyramid Reconstruction Network (InSPyReNet), for HR prediction without any ofHR datasets. We design InSPyReNet to produce a strict image pyramid structureof saliency map, which enables to ensemble multiple results with pyramid-basedimage blending. For HR prediction, we design a pyramid blending method whichsynthesizes two different image pyramids from a pair of LR and HR scale fromthe same image to overcome effective receptive field (ERF) discrepancy. Ourextensive evaluations on public LR and HR SOD benchmarks demonstrate thatInSPyReNet surpasses the State-of-the-Art (SotA) methods on various SOD metricsand boundary accuracy.