How does the model sample frames in every segments to extract features?

[elevanth]

Hi!
I would like to do some experiments with the feature you provided.
What I am confused is, how does the feature extracted?

1. I guess, middle frame in every segment for TSN model for both train and val split, am I right?
2. How does a I3D model extract feature from every segment, considering you shold sample 8 frames in each segment? The sampled frames are continuous or Interval?

[elevanth]

Another question is as follow:
for subaction element:
0LtLS9wROrk_E_002681_002688_A_0000_0006
annotation is:
{'stages': 3,
'timestamps': [[0.04, 3.2], [3.2, 4.49], [4.49, 6.57]]},

there are 3 timestamps in such element, how do you extract ONE feature from these THREE timestamps?