Salient target detection
The purpose of salient object detection is to identify the most eye-catching object from the input image. Simply put, this research direction hopes to identify the subject of the image.
Saliency target detection with deep learning method
Although hand-made features allow traditional salient object detection methods to detect in real time, the shortcomings of these methods limit their ability to detect salient objects in complex scenes. At the same time, the wide application of convolutional neural networks in different image tasks (such as target detection, semantic segmentation, edge detection, etc.) provides new ideas for salient target detection, and has shown surprising surprises in some work The effect is improved. Due to its multi-level and multi-scale features, CNN can accurately capture the most salient regions without using any prior knowledge. In addition, even in the presence of shadows or reflections, multi-level features allow CNN to better locate the boundaries of the detected prominent areas. Due to these advantages, the CNN-based salient target detection method refreshes the historical record on almost all existing data sets, and has become the mainstream method of salient target detection.
Links to related materials
http://mmcheng.net/paperreading/
https://github.com/jiwei0921/SOD-CNNs-based-code-summary-