OCR na prática: Tesseract, PaddleOCR, TrOCR

from paddleocr import PaddleOCR

ocr = PaddleOCR(use_angle_cls=True, lang="pt")

result = ocr.ocr("nota-fiscal.jpg", cls=True)
for line in result[0]:
    box, (text, conf) = line
    print(f"[{conf:.2f}] {text}")

from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
import torch

processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten").to("cuda")

def recognize_line(crop_pil):
    pixel_values = processor(images=crop_pil, return_tensors="pt").pixel_values.to("cuda")
    ids = model.generate(pixel_values, max_length=128)
    return processor.batch_decode(ids, skip_special_tokens=True)[0]

line_crop = Image.open("manuscrito-linha.png")
print(recognize_line(line_crop))

import boto3

textract = boto3.client("textract", region_name="us-east-1")

with open("fatura.pdf", "rb") as f:
    resp = textract.analyze_document(
        Document={"Bytes": f.read()},
        FeatureTypes=["FORMS", "TABLES"],
    )

# Extrair key-value pairs
blocks = resp["Blocks"]
kv = {b["Id"]: b for b in blocks}
for b in blocks:
    if b["BlockType"] == "KEY_VALUE_SET" and "KEY" in b.get("EntityTypes", []):
        # resolver relationships pra achar VALUE correspondente
        pass