Article
Automation & Control Systems
Lu Chen, Panfeng Huang, Yuanhao Li, Zhongjie Meng
Summary: This article proposes an edge-based grasp detection strategy that combines low-level features and a lightweight CNN, introducing two grip criteria and rapidly training a model with limited samples to identify feasible grasps. The method does not require additional sensor information and can work well with only RGB images, outperforming existing grasp search strategies in accuracy and efficiency.
IEEE-ASME TRANSACTIONS ON MECHATRONICS
(2021)
Article
Automation & Control Systems
Hu Cao, Guang Chen, Zhijun Li, Qian Feng, Jianjie Lin, Alois Knoll
Summary: This study proposes an efficient grasp detection network for robotic grasp tasks. The network uses a lightweight generative structure to achieve a balance between high grasp confidence and fast inference speed. It introduces a Gaussian kernel-based grasp representation for encoding training samples and employs receptive field blocks and attention mechanisms for improved feature discriminability and semantic information fusion. Experimental results demonstrate excellent performance on Cornell, Jacquard, and extended OCID grasp datasets with accuracy of 97.8%, 95.6%, and 76.4% respectively.
IEEE-ASME TRANSACTIONS ON MECHATRONICS
(2023)
Article
Computer Science, Artificial Intelligence
Jingjing Li, Wei Ji, Miao Zhang, Yongri Piao, Huchuan Lu, Li Cheng
Summary: Recent years have seen increasing interest in RGB-D Salient Object Detection (SOD) by utilizing depth maps for better distinction of salient objects from complex backgrounds. However, the presence of noise and ambiguity in raw depth images, as well as the coarse object boundaries in saliency predictions, have hindered the progress in this research field. To address these issues, this paper proposes a Depth Calibration and Boundary-aware Fusion (DCBF) framework that incorporates a learning strategy for depth map calibration and a multimodal fusion module for improved object boundary qualities.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Automation & Control Systems
Sheng Yu, Di-Hua Zhai, Yuanqing Xia
Summary: In this article, a novel grasp detection network called efficient grasp detection network (EGNet) is proposed to address the challenges of grasping in stacked scenes. It combines object detection, grasp detection, and manipulation relationship detection tasks. The EGNet adopts the EfficientDet for object detection and modifies some hyperparameters. It introduces a novel grasp detection module that utilizes the feature map from bidirectional feature pyramid network (BiFPN) to output the grasp position and quality score. The EGNet also incorporates manipulation relation analysis and achieves high detection accuracy in different datasets and practical scenarios.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2023)
Article
Computer Science, Artificial Intelligence
Mengke Song, Wenfeng Song, Guowei Yang, Chenglizhao Chen
Summary: Most existing RGB-D salient object detection (SOD) methods focus on cross-modal and cross-level saliency fusion, but their fusion patterns highly depend on the network's adaptability. To overcome this limitation, this paper proposes a Modality-aware Decoder (MaD) that upgrades the decoding process to be modality-aware.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Zhao Zhang, Zheng Lin, Jun Xu, Wen-Da Jin, Shao-Ping Lu, Deng-Ping Fan
Summary: In this paper, a Bilateral Attention Network (BiANet) is proposed for RGB-D salient object detection task, which includes a Bilateral Attention Module (BAM) to better explore salient information in both foreground and background regions, and leveraging multi-scale techniques for improved SOD performance.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Artificial Intelligence
Xiaoqiang Wang, Lei Zhu, Siliang Tang, Huazhu Fu, Ping Li, Fei Wu, Yi Yang, Yueting Zhuang
Summary: This paper proposes a method to enhance RGB-D saliency detection using unlabeled RGB images. A depth decoupling convolutional neural network is used for depth estimation and saliency detection, and a teacher-student framework is introduced for semi-supervised learning, achieving superior performance.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Robotics
Yifei Shi, Zixin Tang, Xiangting Cai, Hongjia Zhang, Dewen Hu, Xin Xu
Summary: Symmetry is commonly found in everyday objects, and humans tend to grasp objects by recognizing symmetric regions. In this research, a learning-based method is proposed to incorporate symmetry into grasp detection, improving the quality of detected grasps. Additionally, a method is introduced to facilitate grasping unseen objects, achieving state-of-the-art performance in experiments.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Jingyu Wu, Fuming Sun, Rui Xu, Jie Meng, Fasheng Wang
Summary: This paper proposes a strategy of aggregation and interaction to extract edge features, depth features, and salient features while maintaining local details and fully extracting global information. By extracting and fusing features in the learning process, it addresses the issues of multi-scale problem and information redundancy, achieving excellent performance in salient object detection.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Information Systems
Tomoki Ikeda, Masaaki Ikehara
Summary: This paper proposes an RGB-D salient object detection method using Saliency and Edge Reverse Attention (SERA), which combines the fusion of saliency and edge features with reverse attention. The method enhances the sharpness of object boundaries and improves the accuracy of detecting important parts of objects.
Article
Computer Science, Artificial Intelligence
Zhiyu Liu, Munawar Hayat, Hong Yang, Duo Peng, Yinjie Lei
Summary: This paper proposes a weakly supervised approach for salient object detection from multi-modal RGB-D data. The approach relies on labels from scribbles, which are easier to annotate compared to dense labels used in conventional fully supervised settings. The design of the approach regularizes the intermediate latent space to enhance discrimination between salient and non-salient objects. Additionally, a contour detection branch and a Cross-Padding Attention Block (CPAB) are introduced to improve semantic boundaries and enhance long-range dependencies among local features. The method outperforms existing weakly supervised methods and is comparable to fully-supervised state-of-the-art models.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Information Systems
Yanan Song, Liang Gao, Xinyu Li, Weiming Shen, Kunkun Peng
Summary: This paper proposes a decoupled single-stage multi-task robotic grasp detection method based on the Faster R-CNN framework, achieving higher object detection accuracy and grasp detection accuracy through the designed network and a new grasp matching strategy.
SCIENCE CHINA-INFORMATION SCIENCES
(2022)
Article
Automation & Control Systems
Ze-Yu Liu, Jian-Wei Liu, Xin Zuo, Ming-Fei Hu
Summary: This paper introduces an iterative refinement architecture that leverages multi-scale features and an attention-based fusion module to address cross-modal fusion and multi-scale refinement in RGB-D salient object detection.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Hongfa Wen, Chenggang Yan, Xiaofei Zhou, Runmin Cong, Yaoqi Sun, Bolun Zheng, Jiyong Zhang, Yongjun Bao, Guiguang Ding
Summary: In this paper, a novel RGB-D saliency model called Dynamic Selective Network (DSNet) is proposed for salient object detection in RGB-D images, leveraging the complementary information between RGB images and depth maps. By optimizing multi-level and multi-scale information and refining boundaries, the proposed DSNet achieves competitive and excellent performance compared to existing state-of-the-art RGB-D SOD models.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Artificial Intelligence
Hao Chen, Youfu Li, Yongjian Deng, Guosheng Lin
Summary: The study aims to develop a systematic solution for RGB-D salient object detection by addressing modal-specific representation learning, complementary cue selection, and cross-modal complement fusion. The proposed hierarchical cross-modal distillation scheme and residual function contribute to learning discriminative features in a new modality.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2021)