Article
Computer Science, Artificial Intelligence
Juliano Emir Nunes Masson, Marcelo Roberto Petry, Daniel Ferreira Coutinho, Leonardo de Mello Honorio
Summary: Multi-View Stereo (MVS) is a critical step in photogrammetry, relying on the ability to match features in different images. Convolutional Neural Networks have been used to solve this problem, but they consume a large amount of Video RAM. This study reduces GPU memory usage and introduces deformable convolutions to improve the performance.
IMAGE AND VISION COMPUTING
(2022)
Article
Chemistry, Multidisciplinary
Qijie Zou, Jie Zhang, Shuang Chen, Bing Gao, Jing Qin, Aotian Dong
Summary: The key to image depth estimation is accurately finding corresponding points between the left and right images. Binocular cameras can directly estimate the depth of the image range, avoiding the need for target recognition accuracy in monocular depth estimation. However, accurately segmenting objects and finding matching points in the ill-posed areas of the left and right images is difficult for binocular stereo matching.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Artificial Intelligence
Boyang Song, Xiaoguang Hu, Jin Xiao, Guofeng Zhang, Tianyou Chen
Summary: In this paper, the authors propose an end-to-end trainable framework called ACINR-MVSNet for multi-view stereo (MVS) with adaptive group-wise correlation and implicit neural depth refinement. The framework consists of a one-stage MVS architecture followed by refinement modules and an implicit neural refinement module. An adaptive group-wise correlation similarity measure is proposed to solve visibility problems, and a pyramid-based feature extraction network is utilized to gather context-aware information. The experiments demonstrate the effectiveness and generalization of the proposed approach.
IMAGE AND VISION COMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Zhi Ling, Kai Yang, Jinlong Li, Yu Zhang, Xiaorong Gao, Lin Luo, Liming Xie
Summary: This paper investigates the inherent factor hindering the adaptive performance of stereo matching networks and proposes a domain-adaptive feature extractor and feature normalization method. Furthermore, the influence of various modules on the performance of the domain-adaptive network is explored.
Article
Chemistry, Multidisciplinary
Xin Ma, Zhicheng Zhang, Danfeng Wang, Yu Luo, Hui Yuan
Summary: In deep learning-based local stereo matching, larger image patches improve accuracy, but unrestricted enlargement leads to saturation. This study proposes an adaptive deconvolution-based disparity matching network by simplifying Siamese convolutional network and adding deconvolution layers, achieving a good trade-off between accuracy and complexity.
APPLIED SCIENCES-BASEL
(2022)
Article
Geosciences, Multidisciplinary
Li Lin, Yuanben Zhang, Zongji Wang, Lili Zhang, Xiongfei Liu, Qianqian Wang
Summary: This paper proposes a satellite image stereo matching network based on attention mechanism to improve the accuracy of stereo matching results. By introducing a new feature extraction module and attention mechanism, this method effectively solves the problems of insufficient surface feature extraction and matching errors. Experimental results demonstrate the superiority of the proposed method in satellite image stereo matching.
FRONTIERS IN EARTH SCIENCE
(2023)
Article
Engineering, Electrical & Electronic
Congxuan Zhang, Junjie Wu, Zhen Chen, Wen Liu, Ming Li, Shaofeng Jiang
Summary: The Dense-CNN stereo matching method proposed in this paper utilizes a novel densely connected network with multiscale convolutional layers and a new loss-function strategy to address image feature loss issues. Experimental results show superior performance in computational accuracy and robustness compared to state-of-the-art approaches.
SIGNAL PROCESSING-IMAGE COMMUNICATION
(2021)
Article
Computer Science, Software Engineering
Jie Liao, Yanping Fu, Qingan Yan, Fei Luo, Chunxia Xiao
Summary: This paper proposes a MVS network for efficient high-resolution depth estimation by adaptively refining and upsampling the depth map to the desired resolution, reducing excessive computation on accurate positions. Experimental results show that the method can generate comparable results with state-of-the-art learning methods, reconstructing more geometric details and consuming less GPU memory.
COMPUTERS & GRAPHICS-UK
(2021)
Article
Computer Science, Information Systems
Ke Zhang, Mengyu Liu, Jinlai Zhang, Zhenbiao Dong
Summary: This paper introduces a new multi-view stereo network with a pyramid attention module to enhance feature representation of Point-MVSNet. Experimental results show that our method outperforms existing state-of-the-art methods on the DTU dataset, demonstrating its effectiveness.
Article
Computer Science, Information Systems
Rafael Weilharter, Friedrich Fraundorfer
Summary: This study introduces an end-to-end deep learning architecture for 3D reconstruction, focusing on reducing memory requirements to utilize information from high-resolution images. By limiting the depth search range and utilizing a pyramid structure to gradually search for depth correspondences, the method can generate highly accurate 3D models using less GPU memory and runtime.
Article
Computer Science, Artificial Intelligence
Chih-Hsuan Huang, Jar-Ferr Yang
Summary: The paper introduces two new stereo matching algorithms that improve the census function and matching aggregation strategy, leading to higher matching accuracy. Experimental results demonstrate that the proposed system outperforms existing stereo matching algorithms in terms of accuracy.
IET COMPUTER VISION
(2022)
Article
Computer Science, Interdisciplinary Applications
Yue Zhang, Chengtao Peng, Ruofeng Tong, Lanfen Lin, Yen-Wei Chen, Qingqing Chen, Hongjie Hu, S. Kevin Zhou
Summary: In this paper, we propose a novel multi-modal tumor segmentation method with deformable feature fusion and uncertain region refinement to address the deficiencies of known methods. Experimental results demonstrate that our method achieves promising tumor segmentation results and outperforms state-of-the-art methods.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2023)
Article
Computer Science, Information Systems
Zhenguo Liu, Zhao Li, Wengang Ao, Shaoshuang Zhang, Wenlong Liu, Yizhi He
Summary: Compared to 3D convolution, 2D convolution is less computationally expensive and faster in stereo matching methods. However, the initial cost volume generated by 2D convolution lacks rich information, resulting in lower robustness and accuracy in the disparity map affected by illumination. To address this, the proposed MCAFNet utilizes multi-scale adaptive cost attention and adaptive fusion to enrich the cost volume. With the improvements, the model achieves better performance in terms of EPE and error matching rates on different datasets.
Article
Computer Science, Artificial Intelligence
Aixin Chong, Hui Yin, Yanting Liu, Jin Wan, Zhihao Liu, Ming Han
Summary: Compared with traditional hand-crafted feature based methods, learning-based stereo matching methods have made significant progress in matching accuracy. However, current CNN-based methods often require a substantial amount of time and memory consumption. To address this issue, we propose an accurate and fast stereo matching network that incorporates multi-hierarchy feature extraction and multi-step cost aggregation. Experimental results demonstrate that our network achieves highly competitive disparity estimation accuracy with fast inference speed.
Article
Optics
Xucheng Wang, Chenning Tao, Zhenrong Zheng
Summary: A two-stage attention-based occlusion-aware light field depth estimation network is proposed in this study, which can achieve accurate depth estimation in occluded regions and ranks first in the 4D light field benchmark.
OPTICS AND LASERS IN ENGINEERING
(2023)