Depth Anything V2

Depth Anything V2 is the state-of-the-art monocular depth estimation model. Trained on 595K synthetic labeled images and 62M+ real unlabeled images, it produces high-quality depth maps from a single RGB image.

Key advantages

1. More fine-grained - better detail than V1
2. More robust - handles challenging scenes
3. 10x faster - than Stable Diffusion-based depth models
4. Lightweight - runs on consumer GPUs
5. Widely used - backbone for FLUX.1 ControlNet Depth

Model variants

Usage

import cv2
import torch
from depth_anything_v2.dpt import DepthAnythingV2
 
model = DepthAnythingV2(encoder='vitl', features=256,
    out_channels=[256, 512, 1024, 1024])
model.load_state_dict(torch.load('depth_anything_v2_vitl.pth'))
model.eval()
 
depth = model.infer_image(cv2.imread('image.jpg'))
# depth is a HxW numpy array

When to use

• 3D reconstruction - generate depth for 3D scenes
• AR/VR - real-world depth sensing
• Robotics - obstacle avoidance
• Image editing - depth-aware adjustments
• ControlNet - conditioning for image generation