RGB-T Saliency Detection Benchmark: Dataset, Baselines, Analysis and a Novel Approach

Unified Information Fusion Network for Multi-Modal RGB-D and RGB-T Salient Object Detection

“…2) RGBT-based SOD Datasets: Moreover, we conduct experiments on three RGB-T SOD datasets, which further demonstrates the validity of our proposed model in multimodal fusion task. VT821 [25] consists of 821 RGB-T image pairs taken by a thermal imager (FLIR A310) and a CCD camera (SONY TD-2073). VT1000 [27] contains 1000 RGB-T image pairs, which are captured by FLIR SC620 camera.…”

“…VT821 Dataset [25] VT1000 Dataset [27] VT5000 Dataset [29] [25] 0.815 0.725 0.462 0.662 0.108 0.836 0.706 0.485 0.715 0.119 0.795 0.680 0.397 0.595 0.114 M3S-NIR [26] 0.859 0.723 0.407 0.734 0.140 0.827 0.726 0.463 0.717 0.145 0.780 0.652 0.327 0.575 0.168 SGDL [27] 0.847 0.765 0.583 0.856 0.787 0.652 0.764 0.090 0.824 0.750 0.558 0.672 0.089 MIED [28] 0 metrics. Especially, from the proposed cross-modal multi-stage fusion module (CMFM), our method achieves impressive gains on NLPR [60] and DUT-RGBD [17], which contain more complicated scenes.…”

“…Recently, more and more works focused on integrating RGB and thermal infrared data together to exploit saliency detection. As the original research, Wang et al [25] built up the first RGB-T SOD dataset and designed a multi-task manifold ranking algorithm for salient object detection. Later, Tu et al [26] utilized a multi-scale cross-modal manifold ranking to integrate different features, and introduced a weight for each modality to solve the ranking model.…”

“…In addition, integrating thermal infrared and RGB data has also shown its effectiveness in salient object detection [25]- [29]. Specifically, thermal infrared data can capture the radiated heat of objects and provide supplementary information for better detection when it comes to cluttered backgrounds or low illumination.…”

RGB-D salient object detection: A survey

“…Therefore, thermal images can provide supplementary information to improve salient object detection when images of salient objects suffer from varying light, glare, or shadows. Some RGB-T models [197][198][199][200][201][202][203][204][205] and datasets (VT821 [199], VT1000 [203], and VT5000 [205]) have already been proposed over the past few years. Like for RGB-D salient object detection, the key aim of RGB-T salient object detection is to fuse RGB and thermal infrared images and exploit the correlations between the two modalities.…”