SBR inference on new video

[8lers]

Hi! Great work on these projects and on this repo. I'm interested in using SBR to predict landmarks from my own video dataset, and had a question about the inference procedure.

Is it possible to get LK smoothed landmarks by extracting individual frames, then running those frames through a pretrained model (e.g., the cpm_vgg16-epoch-049-050.pth snapshot) using the eval.py script? Or is there more that needs to be done to get the smoothed landmarks?