How to run inference using custom images? about nvdiffrec HOT 2 CLOSED

The example provided does not seem to take in a set of images to output a 3D mesh, instead it's a json config file with an hdr input and a obj file (what seems like an existing model?) So I assume what's going on inside is that there are screenshots being taken of this mesh from a set of angles, then asking the model to reconstruct that mesh?

from nvdiffrec.

Hello,
Our method does not rely on a pre-trained model. Please see https://github.com/NVlabs/nvdiffrec/#datasets for a few examples of training with both synthetic images and real photos. Example config: https://github.com/NVlabs/nvdiffrec/blob/main/configs/nerd_gold.json

We currently do not provide detailed instructions on how to run with your own images, but it has been discussed in other issues: #31
#26
#3

from nvdiffrec.