Scribble-based Boundary-aware Network for Weakly Supervised Salient Object Detection in Remote Sensing Images

Abstract

Existing CNNs-based salient object detection (SOD) heavily depends on the large-scale pixel-level annotations, which is labor-intensive, time-consuming, and expensive. By contrast, the sparse annotations (e.g., image-level or scribble) become appealing to the salient object detection community. However, few efforts are devoted to learning salient object detection from sparse annotations, especially in the remote sensing field. In addition, the sparse annotation usually contains scanty information, which makes it challenging to train a well-performing model, resulting in its performance largely lagging behind the fully-supervised models. Although some SOD methods adopt some prior cues (e.g., edges) to improve the detection performance, they usually lack targeted discrimination of object boundaries and thus provide saliency maps with poor boundary localization. To this end, in this paper, we propose a novel weaklysupervised salient object detection framework to predict the saliency of remote sensing images from sparse scribble annotations. To implement it, we first construct the scribble-based remote sensing saliency dataset by relabelling an existing large-scale SOD dataset with scribbles, namely S-EOR dataset. After that,

Category-Aware Saliency Enhance Learning Based on CLIP for Weakly Supervised Salient Object Detection

“…The caption based method [10] uses a caption dataset to train a caption generation task and then transfers language-aligned vision features to generate a language-aware saliency map. The scribble based method [23] uses sparse labels indicating foreground and background to achieve whole object detection for RGB images [27], RGB-D images [1], and remote sensing image [11]. The bounding boxes based method [12] leverages the supervision of bounding boxes to update pseudo labels.…”

“…Most SOD methods follow the fully supervised paradigm, which heavily relies on large datasets of pixel-wise annotations. To reduce the cost of manual labeling, weakly supervised methods that use sparse labels (e.g., image-level category labels [7,8], captions [9,10], scribbles [11], bounding boxes [12], points [13], subitizing [14], etc) have been applied to realize a trade-off between time consumption and performance. Among the weakly supervised signals, the image-level category labels can provide the activation regions related to category labels and help deduce the position of the objects.…”