Depth-Anything-V3-Mono: AI Model Estimates Depth from Single Images

Depth-Anything-V3-Mono is a monocular relative depth estimation model developed by Vufinder on Replicate, built on the Depth Anything 3 framework. It predicts spatially consistent depth from a single RGB image using a plain transformer backbone with a unified depth-ray representation, bypassing complex multi-task learning. The model is trained solely on public academic datasets and claims state-of-the-art performance, outperforming its predecessor Depth Anything 2. Key use cases include robotics navigation, 3D reconstruction, AR/VR scene separation, and depth-guided image segmentation. Notably, the model outputs relative rather than metric depth, meaning real-world scale cannot be determined without additional calibration.