training code

[CleyLyChen]

Hi! Great work!

I noticed that you released training code two weeks ago, but there haven't been updates to the relevant sections in the README. Could you confirm if this training code is usable? If so, could you provide the command used for training on the Epic-Kitchens dataset?

[yuezih]

I have the same request 👀

[gorjanradevski]

Hey @CleyLyChen, @yuezih, thanks for showing interest in our work! :) I'll make sure to release the code relevant for EPIC in a few days (after CVPR deadline, unfortunately)...I'll keep you posted!

[gorjanradevski]

Hey, we've released the multimodal knowledge distillation training code. The only thing remaining to reproduce the experiments in our paper are the datasets (Epic-Kitchens, Something-Something/Else). We store these as large HDF5 files which we cannot upload anywhere (together they're around 1000GB). This gives us two options:

1. We provide you a link from our servers from where you can wget them.
2. We release a script, such that once you download Epic-Kitchens and Something-Something/Else, you can pack them as HDF5 files.

Let us know what do you prefer. Cheers!

[Alice123DH]

Hi! Your work is very nice! I found that there is no dataset. In the two options you provided, I hope that you could provide the script. Because downloading the dataset of 1000GB is too difficult for me. At the same time, I would be very grateful if you could provide me with demos of the dataset to help me reproduce the work and learn from you. Thank you so much for your attention!

[gorjanradevski]

@Alice123DH, @yuezih @CleyLyChen I can forward the datasets (HDF5) only via email using Belnet so please post where should I send you the files.

[gorjanradevski]

Update: I'll update the README by Friday next week with a permanent download link to obtain the HDF5 files for all datasets and modalities we use in the paper.

[gorjanradevski]

We've just updated the README with a download link for the datasets. Closing this issue now.