DistributedSampler

[sakerlrf]

Hello,

When training a model with RawVideoFrames, I need to make sure that there is:

• sufficient shuffling of the data of the data

How could I work this around either on the epoch or the batch level?
Does "DistributedSampler" do such kind of shufflings?

If yes, how can I use this for training data in the main model config script?

Thank you!

[dreamerlin]

In class SampleFrames, we make sure to sufficiently explore and exploit sampling frames