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

Difference between provided checkpoints. #89

Open QwCao opened 4 years ago

QwCao commented 4 years ago

Hi, what is the difference between these three checkpoints, rgb_imagenet, rgb_scratch and rgb_scratch_kin600? Are they all trained on ImageNet first and then on Kinetics?

danielkorat commented 4 years ago

Hi, I'm looking for the answer as well. @QwCao

joaoluiscarreira commented 4 years ago

Hi,

rgb_imagenet --> trained on imagenet then kinetics-400. rgb_scratch --> trained from scratch on kinetics-400. rgb_scratch_kin600 --> trained from scratch on kinetics-600.

Best,

Joao

On Tue, Mar 3, 2020 at 2:23 PM Daniel Korat notifications@github.com wrote:

Hi, I'm looking for the answer as well. @QwCao https://github.com/QwCao

— 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/89?email_source=notifications&email_token=ADXKU2RM6HSQKHYMJFLZDQTRFUHE7A5CNFSM4JVI6MN2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOENTVVZQ#issuecomment-593976038, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADXKU2WYTXG5WJPWN3HIA23RFUHE7ANCNFSM4JVI6MNQ .