prepare_data.py

# Copyright (c) 2020 Sarthak Mittal
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

import argparse
import glob
import os
import pdf2image
import simplejson
import tqdm
import multiprocessing as mp

from invoicenet import FIELDS, FIELD_TYPES
from invoicenet.common import util


def process_file(filename, out_dir, phase, ocr_engine):
    try:
        page = pdf2image.convert_from_path(filename)[0]
        page.save(os.path.join(out_dir, phase, os.path.basename(filename)[:-3] + 'png'))

        height = page.size[1]
        width = page.size[0]

        ngrams = util.create_ngrams(page, height=height, width=width, ocr_engine=ocr_engine)
        for ngram in ngrams:
            if "amount" in ngram["parses"]:
                ngram["parses"]["amount"] = util.normalize(ngram["parses"]["amount"], key="amount")
            if "date" in ngram["parses"]:
                ngram["parses"]["date"] = util.normalize(ngram["parses"]["date"], key="date")

        with open(filename[:-3] + 'json', 'r') as fp:
            labels = simplejson.loads(fp.read())

        fields = {}
        for field in FIELDS:
            if field in labels:
                if FIELDS[field] == FIELD_TYPES["amount"]:
                    fields[field] = util.normalize(labels[field], key="amount")
                elif FIELDS[field] == FIELD_TYPES["date"]:
                    fields[field] = util.normalize(labels[field], key="date")
                else:
                    fields[field] = labels[field]
            else:
                fields[field] = ''

        data = {
            "fields": fields,
            "nGrams": ngrams,
            "height": height,
            "width": width,
            "filename": os.path.abspath(
                os.path.join(out_dir, phase, os.path.basename(filename)[:-3] + 'png'))
        }

        with open(os.path.join(out_dir, phase, os.path.basename(filename)[:-3] + 'json'), 'w') as fp:
            fp.write(simplejson.dumps(data, indent=2))
        return True

    except Exception as exp:
        print("Skipping {} : {}".format(filename, exp))
        return False


def main():
    ap = argparse.ArgumentParser()

    ap.add_argument("--data_dir", type=str, required=True,
                    help="path to directory containing invoice document images")
    ap.add_argument("--out_dir", type=str, default='processed_data/',
                    help="path to save prepared data")
    ap.add_argument("--val_size", type=float, default=0.2,
                    help="validation split ration")
    ap.add_argument("--cores", type=int, help='Number of virtual cores to parallelize over',
                    default=max(1, (mp.cpu_count() - 2) // 2))  # To prevent IPC issues
    ap.add_argument("--ocr_engine", type=str, default='pytesseract',
                    help='OCR used to extract text', choices=['pytesseract', 'aws_textract'])

    args = ap.parse_args()

    os.makedirs(os.path.join(args.out_dir, 'train'), exist_ok=True)
    os.makedirs(os.path.join(args.out_dir, 'val'), exist_ok=True)

    filenames = [os.path.abspath(f) for f in glob.glob(args.data_dir + "**/*.pdf", recursive=True)]

    idx = int(len(filenames) * args.val_size)
    train_files = filenames[idx:]
    val_files = filenames[:idx]

    print("Total: {}".format(len(filenames)))
    print("Training: {}".format(len(train_files)))
    print("Validation: {}".format(len(val_files)))

    for phase, filenames in [('train', train_files), ('val', val_files)]:
        print("Preparing {} data...".format(phase))

        with tqdm.tqdm(total=len(filenames)) as pbar:
            pool = mp.Pool(args.cores)
            for filename in filenames:
                pool.apply_async(process_file, args=(filename, args.out_dir, phase, args.ocr_engine),
                                 callback=lambda _: pbar.update())

            pool.close()
            pool.join()


if __name__ == '__main__':
    main()