Video Classification with TimeDistributed Layer (w/o GPU)

[SujoyKG]

New PR

[fchollet]

Thanks for the PR!

Is there a realistic video classification dataset (e.g. action classification) that you could use? I don't think this model architecture would perform well on a real dataset. In addition, the architecture used seems fundamentally expensive (e.g. use of conv2d instead of separable conv, use of LSTM) which contradicts the claim that it is optimized for CPU. At equal parameter count, much leaner architectures are possible.

I would recommend reworking this into a more conventional video classification example. Please start by finding an appropriate dataset, then we can discuss what architecture we should use and whether to use a pretrained base model.

[SujoyKG]

Thanks for the PR!

Is there a realistic video classification dataset (e.g. action classification) that you could use? I don't think this model architecture would perform well on a real dataset. In addition, the architecture used seems fundamentally expensive (e.g. use of conv2d instead of separable conv, use of LSTM) which contradicts the claim that it is optimized for CPU. At equal parameter count, much leaner architectures are possible.

I would recommend reworking this into a more conventional video classification example. Please start by finding an appropriate dataset, then we can discuss what architecture we should use and whether to use a pretrained base model.

Thanks for the reply. Please note what I am trying to express through this example. The dataset generated is optimized for quick checking of any Video Classification Model; the model is not optimized for CPU, as you have asked. I am more focused to the synthetic dataset generation rather than model building. Before feeding the actual data (real data) to the model, this generated dataset could get feeded to check the model quickly at high level (& without any GPU support).

[SujoyKG]

Would you like to get it added to "Quick Keras Recipes"?

[pcoet]

Thanks for the PR, @SujoyKG. Are you planning to implement this feedback?

I would recommend reworking this into a more conventional video classification example. Please start by finding an appropriate dataset, then we can discuss what architecture we should use and whether to use a pretrained base model.

Let us know if you're still working on this. Otherwise we'll close the request. Thanks!

[SujoyKG]

Hi Team, I really want you to go through the approach & code again. I believe it is valuable & could be accepted. Thanks,

…

-- Sent from mobile

On Tue, 15 Aug, 2023, 8:43 pm David Huntsperger, ***@***.***> wrote: Thanks for the PR, @SujoyKG <https://github.com/SujoyKG>. Are you planning to implement this feedback? I would recommend reworking this into a more conventional video classification example. Please start by finding an appropriate dataset, then we can discuss what architecture we should use and whether to use a pretrained base model. Let us know if you're still working on this. Otherwise we'll close the request. Thanks! — Reply to this email directly, view it on GitHub <#770 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJCGSHJNNFXJBRDFSUCOVODXVOG2PANCNFSM5LSOUBKA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[github-actions]

This PR is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[github-actions]

This PR was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.