PROJECT TITLE :

Salient Object Detection With Lossless Feature Reflection and Weighted Structural Loss

ABSTRACT:

As more and more real-world applications emerge, salient object identification, which tries to identify and find the most prominent pixels or regions in images, is becoming increasingly popular. However, this visual job is extremely difficult, especially in the context of complex images. This research proposes a novel feature learning approach for large-scale object detection based on the intrinsic reflection of natural images. As an example, we use lossless feature reflection to guide the design of a symmetric full convolutional network for learning complementary saliency features. In order to better forecast saliency, the suggested network is supervised by the location, context, and semantic information of relevant items. A novel weighted structural loss function is proposed in order to provide obvious object boundaries and spatial consistency in saliency. Improved performance can be achieved through the use of this structural knowledge. We found that our methodology consistently outperformed the most recent state-of-the-art methods on seven saliency detection datasets, and these results were validated by extensive tests.