search

google-deepmind / kinetics-i3d

Convolutional neural network model for video classification trained on the Kinetics dataset.
Apache License 2.0
1.75k stars 462 forks source link

how to deal with long video and short video #80

Open zouying-sjtu opened 5 years ago

zouying-sjtu commented 5 years ago

i find long video (>>64 frame) may cause out of memary error. short videos also have some problems, the tempral length is too small for tempral pool, which also raise exception.

joaoluiscarreira commented 5 years ago

Hi,

that's right. Solution for long videos: divide into shorter videos, perhaps with a bit of overlap. Solution for short videos: the only option to avoid errors is to change the pooling operation just before logit computation, which uses 'VALID' padding. Changing to 'SAME' padding should prevent the exception.

Joao

On Sun, Oct 6, 2019 at 8:26 AM zouying-sjtu notifications@github.com wrote:

i find long video (>>64 frame) may cause out of memary error. short videos also have some problems, the tempral length is too small for tempral pool, which also raise exception.

— 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/80?email_source=notifications&email_token=ADXKU2TA2HUGJCX54B34DF3QNGHKRA5CNFSM4I5232YKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HP4EXRQ, or mute the thread https://github.com/notifications/unsubscribe-auth/ADXKU2QDEFTTZ47A3URDXNLQNGHKRANCNFSM4I5232YA .