ZoeDepth — Zero-Shot Monocular Depth Estimation 🖼️❓ → 🖼️

▶️ 4.7M runs 📅 Mar 2023 ⚙️ Cog 0.6.1 🔗 GitHub 📄 Paper ⚖️ License
depth-estimation monocular-depth 3d computer-vision

ZoeDepth produces metric depth maps from single images. It works across indoor and outdoor scenes without needing camera calibration or stereo input.

About

ZoeDepth estimates depth from a single image without needing stereo pairs or depth sensors. It produces metric (absolute) depth maps rather than just relative depth, which means the output values correspond to actual distances in meters.

What you can do with ZoeDepth

  • 3D scene understanding — convert flat photos into depth-aware representations
  • Image-to-3D pipelines — use depth maps as input for 3D reconstruction or NeRF workflows
  • Robotics and AR — estimate real-world distances from camera images
  • Video effects — create parallax, bokeh, or fog effects based on scene depth

ZoeDepth combines relative depth estimation with metric depth fine-tuning, giving it strong generalization across indoor and outdoor scenes.

Example Output

Output

Example output

Performance Metrics

8.07s Prediction Time
8.36s Total Time
All Input Parameters
{
  "image": "https://replicate.delivery/pbxt/IPzzqLRb2x6XwGUK28l7dNTFO9MzQG1WmY2sdapZ2tnEdmMF/123.png",
  "model_type": "ZoeD_N"
}
Input Parameters
image (required) Type: string
Input image
model_type Default: ZoeD_N
Output Schema

Output

Type: stringFormat: uri

Version Details
Version ID
6375723d97400d3ac7b88e3022b738bf6f433ae165c4a2acd1955eaa6b8fcb62
Version Created
March 4, 2023
Run on Replicate →