Photorealistic derivative dataset based on olmOCR-mix-1025

Hi allenai team,

I’ve created a photorealistic version of olmOCR-mix-1025:

olmOCR-mix-1025-Photoreal

The pages are enhanced instead of clean digital PDFs, while keeping the original filenames so the existing annotations remain usable.

I hope this can help improve robustness in future olmOCR models.
If anyone needs similar datasets (specific effects or application scenarios), feel free to use it.

Happy to discuss if useful.