Experimental icl

[Jbruner23]

add files and code for experimenting ICL for illustrations
codebase taken from https://github.com/garibida/cross-image-attention

[AmitMY]

Good start! left a few general comments, mostly to organize the files better.

[AmitMY]

Add **/__pycache__ to .gitignore file
then, use git to rm all pycache files from the branch

[AmitMY]

all the code for this implementation should be under video_illustration (the main codebase) then cross_img_icl -> video_illustration/cross_img_icl

That way, each "solution" can be separate, but they can import from eachother

[AmitMY]

all images should be appropriately named (e.g. "two_airplanes_512.jpg")

[Jbruner23]

i need the stem reference

[AmitMY]

if multiple images belong to a "cluster", please create a directory, (for example "airplane") under which you have a "original.jpg", "original_512.png", "mask.png", "mask_512.png"

[AmitMY]

I'll make this comment here for lack of a better place -
under cross_img_icl there should be a README.md file.
It can be as simple as:

# Cross Image ICL (btw i'm not sure what ICL mean, why not Cross Image Attention?)

Implementation copied from https://github.com/garibida/cross-image-attention

Setup and usage are the same as the above repository.

## Example
After setup etc, you can run
....

This is to help reproducibility, and so we can run this on the same files and expect the same result~

[Jbruner23]

the codebase from Cross-Image-Attention paper with small changes. ICL means In Context Learning (which might be the suitable task in computer vision)