An Efficient Monocular Depth Prediction Network Using Coordinate Attention and Feature Fusion
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
The recovery of reasonable depth information from different scenes is a popular topic in the field of computervision. For generating depth maps with better details, we present an efficacious monocular depth predictionframework with coordinate attention and feature fusion. Specifically, the proposed framework containsattention, multi-scale and feature fusion modules. The attention module improves features based on coordinateattention to enhance the predicted effect, whereas the multi-scale module integrates useful low- and high-levelcontextual features with higher resolution. Moreover, we developed a feature fusion module to combine theheterogeneous features to generate high-quality depth outputs. We also designed a hybrid loss function thatmeasures prediction errors from the perspective of depth and scale-invariant gradients, which contribute topreserving rich details. We conducted the experiments on public RGBD datasets, and the evaluation resultsshow that the proposed scheme can considerably enhance the accuracy of depth prediction, achieving 0.051 forlog10 and 0.992 for δ<1.253 on the NYUv2 dataset.
Abstract
The recovery of reasonable depth information from different scenes is a popular topic in the field of computervision. For generating depth maps with better details, we present an efficacious monocular depth predictionframework with coordinate attention and feature fusion. Specifically, the proposed framework containsattention, multi-scale and feature fusion modules. The attention module improves features based on coordinateattention to enhance the predicted effect, whereas the multi-scale module integrates useful low- and high-levelcontextual features with higher resolution. Moreover, we developed a feature fusion module to combine theheterogeneous features to generate high-quality depth outputs. We also designed a hybrid loss function thatmeasures prediction errors from the perspective of depth and scale-invariant gradients, which contribute topreserving rich details. We conducted the experiments on public RGBD datasets, and the evaluation resultsshow that the proposed scheme can considerably enhance the accuracy of depth prediction, achieving 0.051 forlog10 and 0.992 for δ<1.253 on the NYUv2 dataset.
참고문헌 (10)
- W. Zhang, "SEAR : scaling experiences in multi-user augmented reality", IEEE Transactions on Visualization and Computer Graphics, 28(5), pp.1982~1992 (2022)
- S. C. Hsia, "Real-time 2D to 3D image conversion algorithm and VLSI architecture for natural scene", Circuits, Systems, and Signal Processing, 41, pp.4455~4478 (2022)
- J. H. Lee, "Monocular depth estimation using relative depth maps", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.9729~9738 (2019)
- H. Jiang, "High quality monocular depth estimation via a multi-scale network and a detail"preserving objective", Proceedings of 2019 IEEE International Conference on Image Processing(ICIP), pp.1920~1924 (2019)
- L. Wang, "SDC-Depth : semantic divide-and-conquer network for monocular depth estimation", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.538~547 (2020)
- J. Hu, "Revisiting single image depth estimation : toward higher resolution maps with accurate object boundaries", Proceedings of 2019 IEEE Winter Conference on Applications of Computer Vision(WACV), pp.1043~1051 (2019)
- Q. Hou, "Coordinate attention for efficient mobile network design", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.13713~13722 (2021)
- X. Tu, "Efficient monocular depth estimation for edge devices in Internet of Things", IEEE Transactions on Industrial Informatics, 17(4), pp.2821~2832 (2020)
- Xinchen Ye, "DPNet: Detail-preserving network for high quality monocular depth estimation", Pattern Recognition, 109, pp.107578~ (2021), Elsevier BV[DOI]
- "MSFNet: multi-scale features network for monocular depth estimation" (2021)[URL]