94 lines
2.9 KiB
Python
Raw Normal View History

2023-02-03 20:12:03 -08:00
"""Image Reader.
A parser for image files.
"""
import re
from pathlib import Path
2023-02-03 23:38:12 -08:00
from typing import Dict, List, Optional
2023-02-03 20:12:03 -08:00
from gpt_index.readers.base import BaseReader
from gpt_index.readers.schema.base import Document
class ImageReader(BaseReader):
"""Image parser.
Extract text from images using DONUT.
"""
def __init__(self, text_type: str = "plain_text") -> None:
2023-02-03 20:12:03 -08:00
"""Init parser."""
if text_type == "plain_text":
import pytesseract
2023-02-17 19:28:21 -08:00
processor = None
model = pytesseract
else:
2023-02-17 19:28:21 -08:00
from transformers import DonutProcessor, VisionEncoderDecoderModel
processor = DonutProcessor.from_pretrained(
"naver-clova-ix/donut-base-finetuned-cord-v2"
)
model = VisionEncoderDecoderModel.from_pretrained(
"naver-clova-ix/donut-base-finetuned-cord-v2"
2023-02-03 20:12:03 -08:00
)
self.parser_config = {"processor": processor, "model": model}
def load_data(
self, file: Path, extra_info: Optional[Dict] = None
) -> List[Document]:
"""Parse file."""
from PIL import Image
2023-02-17 19:28:21 -08:00
2023-02-03 20:12:03 -08:00
model = self.parser_config["model"]
processor = self.parser_config["processor"]
if processor:
import torch
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
# load document image
image = Image.open(file)
if image.mode != "RGB":
image = image.convert("RGB")
# prepare decoder inputs
task_prompt = "<s_cord-v2>"
decoder_input_ids = processor.tokenizer(
task_prompt, add_special_tokens=False, return_tensors="pt"
).input_ids
pixel_values = processor(image, return_tensors="pt").pixel_values
outputs = model.generate(
pixel_values.to(device),
decoder_input_ids=decoder_input_ids.to(device),
max_length=model.decoder.config.max_position_embeddings,
early_stopping=True,
pad_token_id=processor.tokenizer.pad_token_id,
eos_token_id=processor.tokenizer.eos_token_id,
use_cache=True,
num_beams=1,
bad_words_ids=[[processor.tokenizer.unk_token_id]],
return_dict_in_generate=True,
)
sequence = processor.batch_decode(outputs.sequences)[0]
sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(
processor.tokenizer.pad_token, ""
)
# remove first task start token
text = re.sub(r"<.*?>", "", sequence, count=1).strip()
else:
# load document image
image = Image.open(file)
text = model.image_to_string(image)
2023-02-17 19:28:21 -08:00
return [Document(text, extra_info=extra_info)]