The base ONNX models can be found here.
The scripts for fusing pre/post-processing operations can be found on the conversion_scripts branch.
The models suffixed with _256 take a 256 * 256 sized RGB image as input, whereas other models accept a 512 * 512 sized RGB image.
Read the Depth-Anything-V2 paper