About training data

[rpdahxn]

Hi,

First of all, how awesome your work!

I have a question regarding the training data you used, the OpenVidHD-0.4M dataset. After examining the dataset, I noticed that the volume of clips appears to be quite large, and the dataset seems to cover a wide variety of categories. I was wondering:
1. How many clips did you select or utilize from the OpenVidHD-0.4M dataset for your training process?
2. Did you apply any specific filtering or preprocessing criteria when selecting these clips to ensure consistency across categories?

Thank you !

[encounter1997]

Thank you for your interest! We utilize the complete OpenVidHD-0.4M dataset without any further processing, as we have found that it performs effectively as is.