taoyang1122 / adapt-image-models

[ICLR'23] AIM: Adapting Image Models for Efficient Video Action Recognition
Apache License 2.0
276 stars 21 forks source link

About training time. #23

Closed leexinhao closed 1 year ago

leexinhao commented 1 year ago

In your paper the GPU training time is 15h, which is much smaller than my reproduction (15 x 8 v100 hours). do you multipy the numbers of GPUs? image

taoyang1122 commented 1 year ago

Hi, sorry for the confusion. The time here is the wall-clock time.

007invictus commented 1 year ago

Hi, sorry for the confusion. The time here is the wall-clock time.

when i test the speed of the kinetics400 ,i use num_frames=8 so the memory is about 10000m. but my testing time is 24h x 8v100hours,do you know the possible solution of that issue?