Label frames on one computer, train network on another?

What is the best workflow for labeling images on one computer, and then carrying out training on another computer? I want to label frames on my workstation (no suitable GPU), then train the model using a GPU node on my institution’s HPC cluster.

After labeling frames on my workstation, can I just copy the project folder (which contains the labeled frames & training dataset) to the cluster, and call train_network on the copied version of the config file?

You can port your project anywhere by just moving it, and then only editing the “project_path” in the config.yaml file, see details here:

1 Like