The field of depth estimation in computer vision is rapidly evolving, with a significant shift towards leveraging deep learning (DL) techniques to overcome the limitations of traditional methods. These DL-based approaches are increasingly focusing on improving the accuracy and reliability of depth estimation across various conditions and applications, from 3D reconstruction to robotics. A notable trend is the integration of monocular and stereo depth estimation methods to enhance the robustness and precision of depth maps, especially in challenging scenarios like occlusions and textureless areas. Additionally, there's a growing emphasis on uncertainty quantification and the use of foundation models to improve the reliability and explainability of depth estimation models. The development of novel frameworks and pipelines that combine the strengths of different depth estimation techniques and incorporate advanced features like adaptive disparity selection and confidence-based guidance is pushing the boundaries of what's possible in this field. Furthermore, the empirical comparison of depth sensing cameras and the exploration of new methods for camera pose estimation and focal length recovery are contributing to the advancement of practical applications in robotics and beyond.

Noteworthy Papers

• Relative Pose Estimation through Affine Corrections of Monocular Depth Priors: Introduces solvers for relative pose estimation that account for affine ambiguities, significantly improving over classic keypoint-based solutions.
• Depth and Image Fusion for Road Obstacle Detection Using Stereo Camera: Presents a method combining depth information and video analysis for effective road obstacle detection, demonstrating success in detecting and tracking small objects.
• Empirical Comparison of Four Stereoscopic Depth Sensing Cameras for Robotics Applications: Offers a comprehensive comparison of stereoscopic RGB-D cameras, providing valuable insights for selecting the best camera for specific robotics applications.
• Three-view Focal Length Recovery From Homographies: Proposes a novel approach for recovering focal lengths from three-view homographies, showing faster and more accurate results than existing methods.
• Fixing the Scale and Shift in Monocular Depth For Camera Pose Estimation: Develops a framework for estimating camera pose from point correspondences with associated monocular depths, achieving state-of-the-art results.
• A Critical Synthesis of Uncertainty Quantification and Foundation Models in Monocular Depth Estimation: Fuses uncertainty quantification methods with a state-of-the-art foundation model, laying a foundation for safer and more reliable machine vision systems.
• Predicting Performance of Object Detection Models in Electron Microscopy Using Random Forests: Introduces a random forest regression model for estimating the performance of object detection models in TEM images, providing a robust approach across different datasets.
• MonSter: Marry Monodepth to Stereo Unleashes Power: Combines monocular depth estimation and stereo matching in a dual-branch architecture, achieving top performance across multiple benchmarks.
• StereoGen: High-quality Stereo Image Generation from a Single Image: Proposes a pipeline for generating high-quality stereo images from single images, showing state-of-the-art zero-shot generalization results.
• DEFOM-Stereo: Depth Foundation Model Based Stereo Matching: Incorporates a monocular relative depth model into a stereo-matching framework, demonstrating strong zero-shot generalization and achieving top performance on several benchmarks.