search

google-deepmind / kinetics-i3d

Convolutional neural network model for video classification trained on the Kinetics dataset.
Apache License 2.0
1.74k stars 461 forks source link

about frames #72

Open ghost opened 5 years ago

ghost commented 5 years ago

Hi! Now, I read a I3D paper. But some part is hard to understanding. Is it possible a training frames 64 and test frames 256? I can't understanding. Could you help me please? incep2

joaoluiscarreira commented 5 years ago

Hi Jin-Hyuck,

yes it is possible. These models are convolutional -- you can just slide them over a longer video, then average the multiple temporal predictions at the end.

Best,

Joao

On Tue, Jul 16, 2019 at 9:00 AM Jin-Hyuck Cha notifications@github.com wrote:

Hi! Now, I read a I3D paper. But some part is hard to understanding. Is it possible a training frames 64 and test frames 256? I can't understanding. Could you help me please? [image: incep2] https://user-images.githubusercontent.com/31475037/61276223-91b11c00-a7ea-11e9-9ad1-05aa85c25f88.PNG

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/deepmind/kinetics-i3d/issues/72?email_source=notifications&email_token=ADXKU2TJVCARORCZQNVV7V3P7V53NA5CNFSM4ID6ET6KYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4G7M2TVA, or mute the thread https://github.com/notifications/unsubscribe-auth/ADXKU2XIL3BZ6G6UKD4E7N3P7V53NANCNFSM4ID6ET6A .