The field of computer vision is witnessing significant advancements in depth estimation and object detection, particularly in applications such as autonomous navigation and robotics. Researchers are exploring innovative approaches to improve the accuracy and efficiency of these tasks, including the use of radar-guided polynomial fitting, joint semantic segmentation and depth estimation, and local-global interaction for object detection. These methods are achieving state-of-the-art performance on various benchmark datasets, demonstrating their potential for real-world applications. Notably, some papers are introducing novel architectures and training objectives that preserve structural consistency and enable stable depth density and depth distribution invariant feature extraction. Noteworthy papers include:

• Radar-Guided Polynomial Fitting for Metric Depth Estimation, which achieves state-of-the-art performance on the nuScenes, ZJU-4DRadarCam, and View-of-Delft datasets.
• Vanishing Depth: A Depth Adapter with Positional Depth Encoding for Generalized Image Encoders, which enables stable depth density and depth distribution invariant feature extraction and achieves SOTA results across a spectrum of relevant RGBD downstream tasks.