Depth maps from any camera with Depth Anything in TD

What it does

Monocular depth estimation accelerated by TensorRT, turning a regular camera into a depth source for effects and masking.

How to set it up

1. Install Python 3.11 and CUDA Toolkit 11.8 (tested with TD 2023.11340).
2. Download the repo, set PYTHON_PATH in accelerate.bat and run it to build the TensorRT engine.
3. Load the component in TD and feed it a camera TOP to get a depth map.

Source project: olegchomp/TDDepthAnything

FAQ

Can I share a Python environment with TouchDiffusion?

Yes, the README explains copying the files into the TouchDiffusion folder and running accelerate.bat there.

Which models?

Depth Anything v1 and v2.