Instance-Level Salient Object Segmentation

Abstract

Image saliency detection has recently witnessed rapid progress due to deep convolutional neural networks. However, none of the existing methods is able to identify object instances in the detected salient regions. In this paper, we present a salient instance segmentation method that produces a saliency mask with distinct object instance labels for an input image. Our method consists of three steps, estimating saliency map, detecting salient object contours and identifying salient object instances. For the first two steps, we propose a multiscale saliency refinement network, which generates high-quality salient region masks and salient object contours. Once integrated with multiscale combinatorial grouping and a MAP-based subset optimization framework, our method can generate very promising salient object instance segmentation results. To promote further research and evaluation of salient instance segmentation, we also construct a new database of 1000 images and their pixelwise salient instance annotations. Experimental results demonstrate that our proposed method is capable of achieving state-of-the-art performance on all public benchmarks for salient region detection as well as on our new dataset for salient instance segmentation.

Salient instance segmentation via subitizing and clustering

“…At the clustering step in spectral clustering, we adopt a k-means algorithm that improves the selection of initial clustering centers. The proposed MDNN achieves excellent performance and surpasses all the existing methods when applied to the only available salient instance segmentation dataset (dataset1K) in [18]. Moreover, our saliency model outperforms other state-of-the-art models on public salient object detection datasets.…”

“…This approach promoted the salient objects from region-level to instance-level; thus, each instance could correspond to a bounding box for the first time. Li et al were the first to address the instance-level salient object segmentation task [18]. The developed method used the multiscale refinement network (MSRNet) to produce the saliency maps and contour maps and then used subset optimization to refine the number of proposals.…”

“…However, the previous works like [15]- [17] focused only on the salient regions and overlooked the individual instances of salient regions. Currently, a new challenging task has gradually attracted widespread attention: instance-level salient object segmentation (or salient instance segmentation for short), which was first proposed by et al [18]. Compared to salient object detection, salient instance segmentation attempts to discriminate individual object instances in the salient regions of an image or scene.…”

“…In general, the existing methods locate individual instances in salient regions with the aid of proposals. Li et al combined salient object proposals and contours to generate the final instances [18]. S4Net also uses RPN (Region Proposal Network) to generate a number of proposals, and then uses SID (salient instance discriminator) to segment the salient instances [21].…”

CAGNet: Content-Aware Guidance for Salient Object Detection

“…To obtain multi-scale features, some previous methods adopted parallel networks and fed them with re-scaled images [34] or multi-context super-pixels [28]. Different from these methods, we propose Multi-scale Feature Extraction Module (MFEM) to extract multi-scale features.…”