[Blending] Smoke Randomization Starting Point

[jpennors]

Create a module to get a random dx,dy starting point for smoke addition

[MateoLostanlen]

Hi @jpennors , thnaks for your PR, can you fix the tests ? I'm looking at the code itself but you missing init files to make the package work properly

[MateoLostanlen]

I recommend adding the following code:

from skimage import measure
blobs_labels = measure.label(mask, background=0)
s = [np.sum(blobs_labels==l) for l in np.unique(blobs_labels)[1:]]
mask = blobs_labels==(np.where(s==np.max(s))[0] + 1)

This will improve the mask by removing false detection like text in the corner for example

[jpennors]

Awesome @MateoLostanlen !
I did not know this lalelization package. It's actually working very well.

From :

To :

The code currently keeps the biggest label. We could have done it another way by picking the label which is the closest to the floor (meaning y=0). But I guess this is fine with our use cas 😄

[jpennors]

@MateoLostanlen I got weird failing tests :

• Unable to resolve action gaurav-nelson/github-action-markdown-link-check@main, unable to find version main``
• would reformat syntheticdataset/image_blending.py. I did not update this file, and I don't have any changes locally when running black ./
• ModuleNotFoundError: No module named 'skimage'. It does'nt seem to install scikit-image, but I added in the requirements