Towards the Success Rate of One: Real-time Unconstrained Salient Object Detection
This work addresses efficient and accurate salient object detection for computer vision applications, representing a novel integration of tasks but with incremental methodological improvements.
The paper tackles unconstrained salient object detection by proposing a deep learning approach that directly generates saliency maps with Gaussian distributions, eliminating the need for object proposals and pixel-level annotations, achieving over 100 fps on a single GPU and outperforming existing methods by a large margin.
In this work, we propose an efficient and effective approach for unconstrained salient object detection in images using deep convolutional neural networks. Instead of generating thousands of candidate bounding boxes and refining them, our network directly learns to generate the saliency map containing the exact number of salient objects. During training, we convert the ground-truth rectangular boxes to Gaussian distributions that better capture the ROI regarding individual salient objects. During inference, the network predicts Gaussian distributions centered at salient objects with an appropriate covariance, from which bounding boxes are easily inferred. Notably, our network performs saliency map prediction without pixel-level annotations, salient object detection without object proposals, and salient object subitizing simultaneously, all in a single pass within a unified framework. Extensive experiments show that our approach outperforms existing methods on various datasets by a large margin, and achieves more than 100 fps with VGG16 network on a single GPU during inference.