Self Supervised Depth

Self-supervised depth estimation aims to reconstruct 3D scene depth from monocular video sequences without relying on ground truth depth labels, a significant challenge in computer vision. Current research focuses on improving accuracy and robustness, particularly in dynamic scenes and adverse weather conditions, by employing various architectures including convolutional neural networks (CNNs), transformers, and hybrid models that leverage camera models and geometric constraints like epipolar geometry. These advancements are crucial for applications in robotics, autonomous driving, and medical imaging, where accurate depth perception is essential but labeled data is scarce or expensive to obtain.