[CVPR] BASNet: Boundary-Aware Salient Object Detection

Most of the deep learning methods → focus on the region prediction when it comes to saliency prediction → but now they have created a new loss function where → the BOUNDARY of the object is also considered.

Saliency prediction → we humans are good at this → focusing on the ‘important’ object given an image or video. (but not a lot of methods considers the boundary of the object). Now a lot of deep learning models → combines the different representation of → what this means →

Like a U-net method, → the internal features are used directly as well.

And it seems like they have the right idea → their method really did outperform other SOTA methods. (most of the methods uses FCN → fully convolutional networks → some actually uses RNN networks for iterative predictions).