Speed up image processors - cast to array before BatchFeature

[amyeroberts]

Feature request

Complete work initially started in #28221

Motivation

Converting the model outputs to an array before creating the BatchFeature object can speed up image processing, especially for nested outputs, such as for videos.

Your contribution

Open to the community to add

[yaswanth19]

Hi @amyeroberts , If this Issue is still open then I would like to work on it. If so, can you provide me some more context to work on.

[amyeroberts]

Hi @yaswanth19, apologies for the delay in response.

The idea is similar to what's done here in idefics 3, where the inputs are converted to numpy arrays before passing to BatchFeature.

I'm going to be off for a while, so if you're still keen to take this on, @yonigozlan is the best person to ping for first review 🤗

[csking101]

Hi @amyeroberts @yonigozlan , is this issue still open? If so, could I work on it?

[yaswanth19]

@csking101 I am working on it and will raise the PR soon

[csking101]

@csking101 I am working on it and will raise the PR soon

Sure, thanks for informing me!