This paper proposes a new method for salient object detection by introducing short connections to the skip-layer structures within the HED architecture. The method leverages multi-level and multi-scale features extracted from FCNs to provide more advanced representations at each layer, which is critical for segment detection. The proposed approach achieves state-of-the-art results on five widely tested salient object detection benchmarks, with advantages in terms of efficiency (0.08 seconds per image), effectiveness, and simplicity over existing algorithms. The method also conducts an exhaustive analysis on the role of training data on performance, providing a more reasonable and powerful training set for future research and fair comparisons. The paper discusses the challenges of salient object detection, including the limitations of traditional methods based on hand-crafted features and the need for learning-based methods that can integrate different types of features. It reviews recent deep learning approaches for salient object detection, including CNN-based models and skip-layer structures. The proposed method introduces short connections to the HED architecture to combine features from different levels, allowing deeper side outputs to provide high-level semantic knowledge and shallower side outputs to capture rich spatial information. This combination improves the accuracy and efficiency of salient object detection. The paper also discusses the implementation details of the proposed method, including the use of a fully convolutional network, the introduction of short connections, and the use of a CRF layer for further refining the saliency maps. The method is evaluated on multiple datasets, including MSRA-B, ECSSD, HKU-IS, PASCALS, and SOD, and shows significant improvements in terms of F-measure and MAE scores. The paper also addresses the issue of saliency existence, proposing an additional branch to predict the presence of salient objects in the input image. The proposed method is efficient, with a processing time of 0.08 seconds per image, and achieves state-of-the-art results on multiple benchmarks. The method's effectiveness is demonstrated through extensive experiments and comparisons with existing approaches, showing that it outperforms other methods in terms of accuracy and efficiency. The paper concludes that the proposed method is a promising approach for salient object detection, with potential for further improvements through more advanced models and training data.This paper proposes a new method for salient object detection by introducing short connections to the skip-layer structures within the HED architecture. The method leverages multi-level and multi-scale features extracted from FCNs to provide more advanced representations at each layer, which is critical for segment detection. The proposed approach achieves state-of-the-art results on five widely tested salient object detection benchmarks, with advantages in terms of efficiency (0.08 seconds per image), effectiveness, and simplicity over existing algorithms. The method also conducts an exhaustive analysis on the role of training data on performance, providing a more reasonable and powerful training set for future research and fair comparisons. The paper discusses the challenges of salient object detection, including the limitations of traditional methods based on hand-crafted features and the need for learning-based methods that can integrate different types of features. It reviews recent deep learning approaches for salient object detection, including CNN-based models and skip-layer structures. The proposed method introduces short connections to the HED architecture to combine features from different levels, allowing deeper side outputs to provide high-level semantic knowledge and shallower side outputs to capture rich spatial information. This combination improves the accuracy and efficiency of salient object detection. The paper also discusses the implementation details of the proposed method, including the use of a fully convolutional network, the introduction of short connections, and the use of a CRF layer for further refining the saliency maps. The method is evaluated on multiple datasets, including MSRA-B, ECSSD, HKU-IS, PASCALS, and SOD, and shows significant improvements in terms of F-measure and MAE scores. The paper also addresses the issue of saliency existence, proposing an additional branch to predict the presence of salient objects in the input image. The proposed method is efficient, with a processing time of 0.08 seconds per image, and achieves state-of-the-art results on multiple benchmarks. The method's effectiveness is demonstrated through extensive experiments and comparisons with existing approaches, showing that it outperforms other methods in terms of accuracy and efficiency. The paper concludes that the proposed method is a promising approach for salient object detection, with potential for further improvements through more advanced models and training data.