AlroWilde/olmOCR-mix-1025-Photoreal

⚠️ Important Feedback Invitation
If this dataset has brought stronger positive or negative improvements to your model training,
I warmly welcome you to email me your feedback at any time.
This is very important to me — thank you!
📧Contact: hi@support.alrowilde.com

Photorealistic enhancements of document pages from allenai/olmOCR-mix-1025.
Clean PDF renders are transformed into realistic scanned/photographed images with natural lighting, paper texture, shadows and capture artifacts using Synthetic Engine.

• Image filenames are identical to the original dataset (suffix changed to .jpeg).
• No additional parquet annotation files are provided. Original annotations are plain text (no bbox or line-level polygons). Download them from the original dataset and match by filename.
• Current enhancement effects (new ones will be added continuously):

• Coffee stains 100 pages (2026-07-28)

For requests of new effects or feedback, please leave a comment or email .

Designed for training more robust OCR and document understanding models under real-world conditions.

👤 About Me (Open To Partnerships Global Widely)

🔗 GitHub  |  🌐 Website  |  💼 LinkedIn

This dataset is a derivative work of allenai/olmOCR-mix-1025 (ODC-BY).
Please attribute the original dataset as required.

---