Abstract:
Multi-scale strategies in fully convolutional neural networks (FCNs) improve deep learning-based salient object detection methods. Integrating all scale-specific predictions yields the result.
Multi-scale-based methods have several drawbacks: 1) Directly learning discriminative features and filters to regress high-resolution saliency masks for each scale is difficult; 2) rescaling the multi-scale features could pull in many redundant and inaccurate values, weakening the network’s representational ability.
This paper proposes a residual learning strategy and introduces scale-by-scale coarse prediction refinement. We learn to predict residuals to correct the errors between coarse saliency map and scale-matching ground truth masks.
Dilated Convolutional Pyramid Pooling (DCPP) generates the coarse prediction and guides residual learning through several novel Attentional Residual Modules (ARMs). Residual Refinement Network (R_2_Net) is our name.
We compare the proposed method to five benchmark datasets using state-of-the-art algorithms. Our fully convolutional R_2_Net runs at 33 FPS on one GPU without post-processing.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
