search

jayleicn / VideoLanguageFuturePred

[EMNLP 2020] What is More Likely to Happen Next? Video-and-Language Future Event Prediction
48 stars 4 forks source link

Cannot get the 3D ResNext features of TV clips #4

Closed jianjieluo closed 3 years ago

jianjieluo commented 3 years ago

Hi! Thanks for your great work on VLEP.

I find that we can not extract the 3D ResNext motion features of the tv clips since there are no original tv clips in the released vision data. Therefore, I am wondering whether you have released the 3D ResNext features of TV clips, and if not, will you release them elsewhere in the future?

Thanks a lot!

jayleicn commented 3 years ago

Hi @jianjieluo,

If you do want ResNext motion features, you can actually concat the frames into a video, though at a much lower frame rate, it may still give a reasonable performance. For 3D features of other types, you can find I3D feature at https://github.com/jayleicn/TVRetrieval, SlowFast features at https://github.com/linjieli222/HERO.

Best, Jie

jianjieluo commented 3 years ago

I see. Thanks for your response!