.png){kind=icon}
返回
LFDA: A Framework for Light Field Depth Estimation With Depth Attention
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2024.3393576.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Depth estimation in light field imaging is integral for the accurate rendering of 3D scenes and a crucial task in light field applications. However, the development of a model that simultaneously achieves high accuracy and speed in light field depth estimation remains a significant challenge. Existing networks utilizing dilated convolution achieve state-of-the-art speeds, but they often encounter accuracy limitations, particularly in fine-grained details. In this paper, we introduce a fast and accurate method based on depth-wise cross attention. By integrating cross-attention with existing networks, our approach effectively emphasizes local features, thereby overcoming the accuracy limitations commonly encountered. Our method adopts depth attention to compare the center and side views along the epipolar line. As a result of depth attention, the cost volume was aggregated by similarity information that was based on the attention score. This technique not only maintains computational efficiency but also significantly enhances the performance in fine-grained regions by emphasizing the importance of local feature analysis. We validated the efficacy of depth attention in emphasizing local features. Our experiments were conducted using the 4D HCI Benchmark, employing evaluation metrics such as BadPixel and MSE. The results demonstrate remarkable performance in estimating fine depth changes, primarily due to the focus on local features, thereby offering a balanced solution in terms of both speed and accuracy. The code is available: https://github.com/syt06007/LFDA.
Keyword:
Costs
Estimation
Convolution
Feature extraction
Three-dimensional displays
Vectors
Light field
depth estimation
attention
deep learning
期刊
IF:
3.6
论文数:
9.8W
被引数:
29.4W
机构
引用论文
分享收藏OMERACT agreement and reliability study of ultrasonographic elementary lesions in osteoarthritis of the foot.png)
RMD Open
OAAI
RMD OpenIF0
OAAI分享收藏Research on the Efficiency of Wireless Power Transfer System Based on Multi-Auxiliary Transmitting Coils基于多辅助发射线圈的无线电能传输系统效率研究
PREAI
分享收藏
查看更多内容