Closed jalaxy33 closed 4 years ago
Yes, these numbers (0, 7, 9, 12, 21...) are the type_idx
where you can find in the csv file. They are official definition and should be used for the official evaluation code.
https://github.com/HYPJUDY/Decouple-SSAD/blob/e3b65392cc4be7dd3887bd38996792ce6dc698f0/data/gen_thumos14_anno.py#L21-L25
https://github.com/HYPJUDY/Decouple-SSAD/blob/e3b65392cc4be7dd3887bd38996792ce6dc698f0/config.py#L31-L32
Thanks for your quick reply. Now what I truly want to know is how to decide the number tagged to each label? For example, I want to know why 7 is tagged to 'BaseballPitch' and 9 is tagged to 'BasketballDunk'. Is there any rule on this, or I could arbitrarily tag different number to each label?
This is the official definition. There are 101 classes for UCF101 dataset indexed in alphabetical order: List of action classes and their numerical index THUMOS14 uses a subset of 20 classes of them.
OK, I see. Thanks for explaining.
what does each number in the list stand for in the gen_thumos14_anno.py. Is it the same as the "type_idx" params in this csv file?