search

ojedaf / vCLIMB_Benchmark

27 stars 3 forks source link

pkl file #1

Closed fangyaoyu1999 closed 2 years ago

fangyaoyu1999 commented 2 years ago

I would like to ask how the pkl file is generated?

ojedaf commented 2 years ago

For the pkl files that contain the tasks, we divide the datasets into 10 and 20 tasks without overlap. They are generated randomly, taking into account the Train, Val, and Test partitions of each dataset. It is to simulate a real scenario when you do not know when and what classes will appear. We provide all the partitions to maintain a fair comparison. I appreciate your interest :)

fangyaoyu1999 commented 2 years ago

If so, can you send me a copy of this randomly generated code? I want to try it on other datasets like hmdb51 and ssv1

YeungChiu commented 1 year ago

I would like to ask how the pkl file is generated?

Perhaps the author wants you to make the pkl files for yourself?

ojedaf commented 1 year ago

@YeungChiu and @fangyaoyu1999, you can do the pkl files by yourself if you want. However, the idea is to have standardized tasks saved in pkl files to maintain the same scenarios for all works (Better comparison). As I mentioned, the tasks on the pkl files were randomly selected without overlap, considering all classes in the desired dataset. In what dataset do you want to work? Thank you very much for your interest in our work.