search

Phoenix1327 / tea-action-recognition

The Pytorch code of the TEA module (Temporal Excitation and Aggregation for Action Recognition)
191 stars 31 forks source link

Sampling method when training #9

Closed AliceCrystal closed 3 years ago

AliceCrystal commented 3 years ago

Thanks for your greate job! For training Kinetics400, is the sampling method normal sample or dense sample?

Phoenix1327 commented 3 years ago

normal sample

AliceCrystal commented 3 years ago

If my understanding of the paper and source code is correct, the test uses a dense sample. so would a normal sample for training and a dense sample for testing be inconsistent?