WARAJA/Tzefa-Word-OCR-TrOCR

Tzefa Word OCR Model (Fine-tuned TrOCR)

Fine-tuned TrOCR model for recognizing individual handwritten Tzefa keywords.

Architecture

• Base model: microsoft/trocr-small-stage1
• Fine-tuned on: Handwritten Tzefa keywords (uppercase Hebrew programming commands, number words, variable names)
• Framework: HuggingFace Transformers (VisionEncoderDecoderModel)

Vocabulary

Tzefa keywords include:

• Commands: MAKEINTEGER, MAKESTR, MAKELIST, MAKEBOOL, PRINTINTEGER, PRINTSTR, ADD, SUBTRACT, MULTIPLY, DIVIDE, WHILETRUE, IFTRUE, etc.
• Number words: ZERO through ONEHUNDRED
• User-defined variable names

Usage

from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image

# use_fast=False is required to prevent tokenizer conversion crash
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-small-stage1", use_fast=False)
model = VisionEncoderDecoderModel.from_pretrained("WARAJA/Tzefa-Word-OCR-TrOCR")

image = Image.open("word_crop.png").convert("RGB")
pixel_values = processor(image, return_tensors="pt").pixel_values

generated_ids = model.generate(pixel_values)
text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(text)

Related

• Full Tzefa Pipeline
• OCR Demo
• Word OCR Dataset