search

google-deepmind / kinetics-i3d

Convolutional neural network model for video classification trained on the Kinetics dataset.
Apache License 2.0
1.75k stars 464 forks source link

How can I train my own model? #22

Open DeepStillWater opened 6 years ago

DeepStillWater commented 6 years ago

How can I train my own model?

BaranovMykola commented 6 years ago

Same question

joaoluiscarreira commented 6 years ago

Hi, you can train the model in the usual way. Try looking at some image classification training examples. The steps are as follows: 1. Get training data (videos + ground truth), 2. Load pre-trained I3D model weights, 3. optimize them using SGD to fit to your data.

The only difference to image classification example code available on the web is that you'll be loading video data, typically sets of 64-frame clips. You may want to create some preprocessing code if you want to do data augmentation.

Joao

On Thu, Sep 20, 2018 at 8:32 AM Mykola Baranov notifications@github.com wrote:

Same question

— 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/22#issuecomment-423072668, or mute the thread https://github.com/notifications/unsubscribe-auth/AO6qaj-m8fdygqNzu6ar1Cqgz75qRj8nks5uc0SlgaJpZM4Tf0Xl .

thanhnt1995 commented 6 years ago

Fucking answer!

thanhnt1995 commented 6 years ago

We want to know more clearly!!!

thanhnt1995 commented 6 years ago

Ex your training API

xyy304519983 commented 5 years ago

Hi, you can train the model in the usual way. Try looking at some image classification training examples. The steps are as follows: 1. Get training data (videos + ground truth), 2. Load pre-trained I3D model weights, 3. optimize them using SGD to fit to your data. The only difference to image classification example code available on the web is that you'll be loading video data, typically sets of 64-frame clips. You may want to create some preprocessing code if you want to do data augmentation. Joao … On Thu, Sep 20, 2018 at 8:32 AM Mykola Baranov @.***> wrote: Same question — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#22 (comment)>, or mute the thread https://github.com/notifications/unsubscribe-auth/AO6qaj-m8fdygqNzu6ar1Cqgz75qRj8nks5uc0SlgaJpZM4Tf0Xl .

Hi, could you please explain more in detail? Thanks.