FAQ - MichiganCOG/ViP GitHub Wiki

Q: How do I use the preprocessing transforms?

A: They should be used in the model class. Examples can be seen in existing model.py files. Additional examples for each preprocessing transforms can be found in the test class at the bottom of ViP/datasets/preprocessing_transforms.py.

Q: What are the JSON files for?

A: They contain all annotations and paths to video frame images. These files must be set up a specific way according to our ViP/datasets/templates/. This is so that ViP can automatically extract clips and load the data from the JSON file into your dataset's __get_item__ function.

Q: How do I load a model trained in ViP?

A: Use pretrained in your config.yaml or as an argument and point it to the saved pickle checkpoint in ViP/results. Ex: --pretrained /ViP/results/c3d/exp1/checkpoints/HMDB51_epoch20.pkl