[Docs] simplest way to check performance of mmaction2 with pretrained model

[sourabCV]

The doc issue

I want to test the inference of mmaction2 with a pre-trained model in python. I have a 4 minute video with multiple actions I want to pass that in mmaction2 to generate output and save that video as output.mp4 with actions name. But unable to find documentation that shows my requirements.

Suggest a potential alternative/fix

pls someone add a simpler implementation code with a pre-train model in Python

[gguzzy]

Hello dude, as far as I understood, you want to use pretrained model weights (which you can find from the model zoo regarding the selected model chosen for inference), and according to this, you want to try the inference by running the scripts:

• demo.py
• demo_skeleton.py
• demo_spatiotemporal_det.py

An example of running one of the above script 'https://github.com/open-mmlab/mmaction2/blob/main/demo/README.md'

Anyways, I opened an issue as well, regarding in general on how to train a customized dataset, and how to evaluate their evaluation performances based on the metrics already used.

I was able to run the whole process, but the point of opening that is to show how to really be able to use such a powerful framework, I hope they will understand, and I hope I can publish a better documentation for those.

Thanks.