Customized dataset feature extraction in Charades-STA style

[xijun-cs]

Hi,

Thanks for your awesome work!

I wanna try your method on a customized dataset, which is similar to Charades-STA.

How can I extract the Slowfast + CLIP feature and the text feature on my own?

Best

[wjun0830]

Hello.

Thank you for your interest in our work. For feature extraction, we suggest you to take a look at Official github repo for CLIP and Slowfast!

Thanks

[xijun-cs]

Thanks for your reply.

For double-check, is the clip model ViT-B/32 and slow-fast model SLOWFAST_8x8_R50?

Thanks

[ArlixLin]

Thanks for your reply.

For double-check, is the clip model ViT-B/32 and slow-fast model SLOWFAST_8x8_R50?

Thanks

I want to know it too~

[wjun0830]

You might want to check out issue https://github.com/jayleicn/moment_detr/issues/19

[wjun0830]

And for the clip model yes we used clip b 32 model

[ArlixLin]

Thanks, I found my way (ORZ): Your job ->Umt->Moment-DETR->HERO_Video_Feature_Extractor

[tiesanguaixia]

And for the clip model yes we used clip b 32 model

Hi, the visual and text encoder are frozen, just use the offline features, right? Thank you!

[wjun0830]

And for the clip model yes we used clip b 32 model

Hi, the visual and text encoder are frozen, just use the offline features, right? Thank you!

Yes you are right