InternVideo icon indicating copy to clipboard operation
InternVideo copied to clipboard

why InternVid-200M performance is not as good as ViClip-InternVid-10M-FLT.pth

Open dragen1860 opened this issue 1 year ago • 1 comments

hi, since 200M pretrained dataset is much bigger than 10M version, so why the zero shot performance is not superior than 10M?

dragen1860 avatar Feb 02 '24 04:02 dragen1860

i guess it's because 10m is the subset that somehow max the data diversity

Todibo99 avatar Feb 29 '24 09:02 Todibo99