SalFBNet: Learning pseudo-saliency distribution via feedback convolutional networks

Feed-forward only convolutional neural networks (CNNs) may ignore intrinsic relationships and potential benefits of feedback connections in vision tasks such as saliency detection, despite their significant representation capabilities. In this work, we propose a feedback-recursive framework (SalFBNet) for detection. The proposed model can learn abundant contextual representations by bridging recursive pathway from higher-level feature blocks to low-level layers. Moreover, create large-scale Pseudo-Saliency dataset alleviate the problem data deficiency We first use distribution pseudo-ground-truth. Afterwards, fine-tune on existing eye-fixation datasets. Furthermore, present novel Selective Fixation Non-Fixation Error (sFNE) loss facilitate better distinguishable eye-fixation-based features. Extensive experimental results show that our SalFBNet with fewer parameters achieves competitive public detection benchmarks, which demonstrate effectiveness data.