Hello great minds,
I need some advice about my dataset/dataloader. Any comments are welcomed!
I have 100 videos.
Every video has around 300 frames.
Every video has its own json file including frame number, and .
I would like the dataset/dataloader:
for each video:
- Take n frames(adjacent) (eg. frames[0,n], frames[1,n+1], …, frames[end-n, end])
- Check if each of the n frames meet and request, if yes, take this n frames as one sample(input) , if no, ignore it and try with next n frames.
Continues with other 99 videos.
When loading the dataset, I could either
- Load two samples from the same video
- Load two samples from different videos
- Just load one sample
It is a mess in my brain about how to arrange the code. Please offer me some advice. Thank you in advance!!!