Depth Anything

Depth Anything

Depth Anything is a foundation model for monocular depth estimation trained on an unprecedented 62 million diverse images. It outperforms previous specialised models across all benchmarks while being 10× faster than competing approaches, enabling robust single-image depth perception for robotics, AR, and 3D reconstruction.

Key Features

• Training on 62M images (licensed + pseudo-labelled unlabelled data) for unmatched generalisation
• Metric depth variant (Depth Anything V2 Metric) for robotics and AR with absolute scale
• Three model sizes (Small/Base/Large) for edge-to-cloud deployment
• Video depth stabilisation with temporal consistency
• HuggingFace Transformers integration for one-line inference

Quick Start

pip install transformers torch pillow

from transformers import pipeline
from PIL import Image

pipe = pipeline(
    task="depth-estimation",
    model="depth-anything/Depth-Anything-V2-Small-hf",
)
image = Image.open("image.jpg")
depth = pipe(image)["depth"]
depth.save("depth.png")

npx ai-supply add depth-anything-monocular-depth

Curated mirror of the open-source Depth Anything (Apache-2.0). Get it from the source.