Related Experiment Video
Updated: Jun 11, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
478
LapUNet: a novel approach to monocular depth estimation using dynamic laplacian residual U-shape networks
Yanhui Xi1,2, Sai Li3,4, Zhikang Xu3,4
1School of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha, 410114, Hunan, China. xiyanhui@126.com.
Scientific Reports
|October 9, 2024
Summary
LapUNet enhances monocular depth estimation by introducing a novel dynamic Laplacian residual U-shape module. This method improves depth map detail and edge clarity, outperforming existing models on benchmark datasets.
Area of Science:
- Computer Vision
- Deep Learning
Background:
- Monocular depth estimation is crucial but challenging.
- Existing encoder-decoder models struggle with detail and edge clarity due to simple upsampling.
Purpose of the Study:
- To introduce LapUNet, a novel network for improved monocular depth estimation.
- To address the limitations of current methods in capturing structural details and sharp edges.
Main Methods:
- The proposed LapUNet utilizes a ResNeXt101 encoder and a decoder with a novel dynamic Laplacian residual U-shape (DLRU) module.
- The DLRU module fuses dynamic Laplacian residuals during upsampling for high-frequency feature enhancement.
- Atrous spatial pyramid pooling (ASPP) and a depth map fusion module are incorporated for multi-scale context and feature integration.
Main Results:
- LapUNet demonstrates superior performance compared to previous methods on the KITTI and NYU Depth V2 datasets.
- The model achieves a good balance between performance and moderate model size.
Conclusions:
- The proposed LapUNet effectively improves monocular depth estimation by enhancing structural details and edge clarity.
- The method's effectiveness is validated through 3D reconstruction and target ranging applications.
