How to use PAN for custom action recognition datasets?

[monacv]

Thanks a lot for providing the code and documentation. Could you please provide some instructions on how I could use this for my own custom dataset that has 20 action classes?

[Fritskee]

go to ops/dataset_config.py and add the specifics of your dataset there in the exact same way as is done for the other datasets.

Then, at the bottom of dataset_config.py, in def return_dataset, you add the name of your dataset to the dict_single.

Besides this, you need to make sure that you have the train/test/val text files in the same format as is mentioned in the ReadMe (you can check the temporal-shift-module github repo if you cannot find it here). And you'll also need a category.txt file