Depth Anything V2 Draws Fresh Attention for Single-Image Depth Estimation

Depth Anything V2, a monocular depth estimation model that predicts the distance to each point from a single RGB image, is drawing renewed attention within the developer community. Published as a paper on arXiv in June 2024 and distributed with code and models on GitHub, the model was accepted to NeurIPS 2024. Its defining feature is the ability to generate a depth map from a single ordinary camera image without dedicated sensors such as stereo cameras or LiDAR.