In my case (for Wildlife Insights), I want to resize images before calling process_image(). Would you like to support this use case in the main repo through a new parameter?
Looks non-controversial, and if we can support Wildlife Insights without forking little bits of inference code, that's a win/win. I confirmed the following in testing:
This change has no impact by default
If I hard-code skip_image_resize to "True" and supply random images, most images fail (this is WAI)
If I hard-code skip_image_resize to "True" and supply images already resized to 1280, I get the same results as I do with skip_image_resize==False
If I hard-code skip_image_resize to "True" and supply images already resized to some other size that's still a stride boundary, things mostly don't fail, but I get slightly different results (still reasonable, but different)
Hi Dan,
In my case (for Wildlife Insights), I want to resize images before calling
process_image()
. Would you like to support this use case in the main repo through a new parameter?Ștefan