Add files via upload

[vijayiyer05]

Update & streamline data streaming logic: implment simple/direct streaming & stub code for later datastore implementaiton

[vijayiyer05]

This is what I got ready for today's workshop. This re-implements the data streaming/framing logic from the current version from last year.

Feedback requested & appreciated on this interim state.

Two things remain:

1. Adding an implementation that configures and uses the deepinterpolation datastore
2. Resolving the start frame for the inference loop. Afterwards, @ramnarak & I agreed that there doesn't seem to be a good rationale for why the main predict call started at frame 30. As shown below. We're curious to get your opinion @tkuenzmw, in case we're missing something?

For #2, my instinct is to either run from the begining or to give a range of allowable start frames for the inference "streaming" loop.

[tkuenzmw]

2. Resolving the start frame for the inference loop. Afterwards, @ramnarak & I agreed that there doesn't seem to be a good rationale for why the main predict call started at frame 30. As shown below. We're curious to get your opinion @tkuenzmw, in case we're missing something?

Well it is my understanding that the networks needs 60 frames as input and will generate the "missing" center frame between 1-30 and 31-60, right? If you start at frame 1 of the stack, are you using frames 1-60 as input?
Am I missing something obvious?

[stevevanhooser]

That is my recollection. There won't be output from the first 30 frames because you need 30 frames in the past to predict the center.