create a copy of labelled data

[Chrysovalantis000]

Hello,

I have labelled more than 1000 frames (one project). I want to copy this project and with a subset of those frames train a second model.
Is there any way to create a duplicate of the first project?

Chrysovalantis

[themattinthehatt]

I would recommend the following workflow for this:

1. In your original project, with all labeled data, you will find a file named .../Pose-app/data/<project_name>/CollectedData.csv. This csv file contains all the ground truth labels, as well as the relative paths of the image files (in the first column). Make a copy of this file, say .../Pose-app/data/<project_name>/CollectedData_subset1.csv, and delete the rows of the frames you do not want included in this subset.
2. Before training a model (before or after launching the app, doesn't matter), open the file .../Pose-app/data/<project_name>/model_config_<project_name>.yaml. Inside you'll find a top-level heading named "data", under which there is a field named "csv_file". You should see the value set to "CollectedData.csv". This tells the model which file to look at for the training data. If you change this to "CollectedData_subset1.csv" and save the file, whenever you go to train the next model it will read this csv file instead of the original one.
3. After your model is done training, I would recommend renaming the folder from "HH-MM-SS_super-0" (for example) to "HH-MM-SS_subset1_super-0" or something similar, to keep track of what subset of data this particular model was trained on. If you accidentally forget, have no fear, when you open up any model folder you will see a "config.yaml" file inside, which is a copy of the config file used to train that particular model. So you can also look inside these and see what the value of the "data.csv_file" field is.

Apologies for the hacky solution, but I think that should work for you!