Bluesky Thread DeepSeek-OCR on handwriting October 20, 2025 View original thread DeepSeek-OCR on handwriting still not as good as a pharmacist 22 fr though, this is really good 5
![Here’s the document faithfully transcribed and formatted in Markdown:
⸻
DEEPSEEK-OCR 3B (MoE)
Original Image (Left) Extracted OCR Text (Right)
Name: Armando Cozain <input(partial)/ocr/input>[158, 89, 505, 108]</input>Name: Armando Cozain
Address: West Kondo, Makati City <input(partial)/ocr/input>[158, 108, 508, 127]</input>Address: West Kondo, Makati City
Age: 29 Sex: M Date: 14-03-90 <input(partial)/ocr/input>[158, 128, 505, 150]</input>Age: 29 Sex: M Date: 14-03-90
Rx: (Heniox) <input(partial)/ocr/input>[158, 175, 505, 200]</input>(Heniox)
Amoxicillin 500 mg Cap #21 <input(partial)/ocr/input>[158, 200, 506, 220]</input>Amoxicillin 500 mg Cap #21
Sig: 1 cap 2× a day for seven days <input(partial)/ocr/input>[158, 230, 506, 254]</input>Sig: 1 cap 2× a day for seven days
Physician’s Sig.: Lala Cruz <input(partial)/ocr/input>[158, 274, 505, 297]</input>Physician’s Sig.: Lala Cruz
Lic. No.: 1234567 PTR No.: 1234567 <input(partial)/ocr/input>[158, 298, 505, 318]</input>Lic. No.: 1234567 PTR No.: 1234567
(Signature Block) <input(partial)/ocr/input>[158, 330, 505, 350]</input>(Signature Block)
⸻
Summary
This figure shows DeepSeek-OCR 3B (Mixture-of-Experts) extracting structured text data from a handwritten or printed medical prescription.
• Left: Original image annotated with bounding boxes for each detected field.
• Right: Model output displaying the recognized text with bounding-box coordinates for each detected region.
• Demonstrates DeepSeek-OCR’s precision in both text transcription and spatial layout mapping from document images.
⸻](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ckaz32jwl6t2cno6fmuw2nhn/bafkreih6jlk4lggy2n5lfggi6ym5dfuasftbciuzqjbytclb3yvits6iji@jpeg)