@LinasKo I see a lot of copy to and copy operations. Have you measured how those impact performance?

@LinasKo I see a lot of copy to and copy operations. Have you measured how those impact performance?

No. I'll do that, will let you know the results. If we prefer to optimize, I can do that.

1. Renamed to ensure_cv2_image_... and ensure_pil_image_.... I've seen this before, and we're not necessarily doing a conversion.
2. Removed a few copy operations. I finally understand why cv2 has the destination parameter in methods like addWeighted, despite returning an image :)

The Colab now has tests in the end, comparing the performance on a 10,000x10,000 size empty image.
Results should show performance dips if any copying is expensive.

With the latest version, performance is equivalent or slightly better than in base supervision.

Hey @onuralpszr ,

Given you've worked on it before, would you mind reviewing these changes too? There's a colab that shows how it works, only requiring that you upload the font file.

@LinasKo start reviewing