Closed yuchen-ji closed 1 year ago
Hi, @yuchen-ji , The specific number of videos used for training may depends on your data quality and action categories. Generally speaking, I think you can try to train your skeleton action recognition model after collecting more than 10 samples per action category.
Thank you so much. Generally speaking, do you think it is suitable to set 100 sets of videos for each action, when action recognition is done from similar perspectives?
100 videos / action is sufficient for training I guess.
Thank you for your suggestion. I will first train with 100 videos per action.
Thanks for your solid work. If I want to collect a custom dataset, approximately including less than 10 types of simple actions. How much video data needs to be collected for each action?