About 3D Patch Linear

[JUNJIE99]

Hi! Thank you for your great work!

I came across your paper where you mentioned that "3D linear generates worse results than 2D linear...., and will pretrain on a large video-text dataset to unleash its potential in future work."

I'm curious to know if you have conducted the experiment as mentioned and if there are any further results available. If this experiment has been attempted, I'm particularly interested in understanding the scale of the pretraining dataset required for 3D linear to surpass 2D linear performance.

Looking forward to learning more about your findings.