Unlocking Self-Supervised Monocular Depth Estimation with Diverse YouTube Datasets
Self-supervised learning with diverse datasets from YouTube enables zero-shot generalization in monocular depth estimation, outperforming existing approaches and even supervised methods.
EC-Depth: Exploring Self-supervised Monocular Depth Estimation in Challenging Scenarios
EC-Depth introduces a novel two-stage framework for robust depth estimation in challenging scenarios.
Enhancing Monocular Depth Estimation through Strengthened Pose Information
Introducing SPIdepth, a novel self-supervised approach that significantly improves monocular depth estimation by focusing on the refinement of the pose network, leading to substantial advancements in depth prediction accuracy.
Manydepth2: A Motion-Aware Self-Supervised Approach for Monocular Depth Estimation in Dynamic Scenes
Manydepth2 leverages optical flow and coarse depth information to construct a motion-guided cost volume, enabling precise depth estimation for both dynamic objects and static backgrounds in an efficient manner.
Self-Supervised Monocular Depth Estimation with Large Kernel Attention for Improved Accuracy and Detail Recovery
A self-supervised monocular depth estimation network that utilizes large kernel attention to model long-distance dependencies while maintaining feature channel adaptivity, and an upsampling module to accurately recover fine details in the depth map, achieving competitive performance on the KITTI dataset.
CCDepth: A Lightweight Self-supervised Depth Estimation Network with Enhanced Interpretability
The proposed CCDepth network leverages convolutional neural networks (CNNs) and the white-box CRATE transformer to efficiently extract local and global features, enabling lightweight and interpretable depth estimation.