kaldi_data_generator.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import argparse
import logging
import os
import random
from enum import Enum
from io import BytesIO
from pathlib import Path
from typing import Tuple
import time
import cv2
import numpy as np
import requests
import tqdm
from PIL import Image
from apistar.exceptions import ErrorResponse
from arkindex import ArkindexClient, options_from_env

Box = Tuple[int, int, int, int]

logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s %(levelname)s/%(name)s: %(message)s"
)
logger = logging.getLogger(os.path.basename(__file__))

api_client = ArkindexClient(**options_from_env())

SEED = 42
random.seed(SEED)


def download_image(url):
    '''
    Download an image and open it with Pillow
    '''
    assert url.startswith('http'), 'Image URL must be HTTP(S)'
    # Download the image
    # Cannot use stream=True as urllib's responses do not support the seek(int) method,
    # which is explicitly required by Image.open on file-like objects
    resp = requests.get(url)
    resp.raise_for_status()

    # Preprocess the image and prepare it for classification
    image = Image.open(BytesIO(resp.content))
    logger.debug('Downloaded image {} - size={}x{}'.format(url,
                                                           image.size[0],
                                                           image.size[1]))

    return image


def write_file(file_name, content):
    with open(file_name, 'w') as f:
        f.write(content)


class Extraction(Enum):
    boundingRect: int = 0
    polygon: int = 1


class KaldiDataGenerator:

    def __init__(self, dataset_name='foo', out_dir_base='/tmp/kaldi_data', grayscale=True,
                 extraction=Extraction.boundingRect, accepted_slugs=None, accepted_classes=None, filter_printed=False):
        self.out_dir_base = out_dir_base
        self.dataset_name = dataset_name
        self.grayscale = grayscale
        self.extraction_mode = extraction
        self.accepted_slugs = accepted_slugs
        self.should_filter_by_slug = bool(self.accepted_slugs)
        self.accepted_classes = accepted_classes
        self.should_filter_by_class = bool(self.accepted_classes)
        self.should_filter_printed = filter_printed
        self.out_line_text_dir = os.path.join(self.out_dir_base, 'Transcriptions', self.dataset_name)
        os.makedirs(self.out_line_text_dir, exist_ok=True)
        self.out_line_img_dir = os.path.join(self.out_dir_base, 'Lines', self.dataset_name)
        os.makedirs(self.out_line_img_dir, exist_ok=True)

    def get_image(self, image_url: str, page_id: str) -> 'np.ndarray':
        out_full_img_dir = os.path.join(self.out_dir_base, 'full', page_id)
        os.makedirs(out_full_img_dir, exist_ok=True)
        out_full_img_path = os.path.join(out_full_img_dir, 'full.jpg')
        if self.grayscale:
            download_image(image_url).convert('L').save(
                out_full_img_path, format='jpeg')
            img = cv2.imread(out_full_img_path, cv2.IMREAD_GRAYSCALE)
        else:
            download_image(image_url).save(
                out_full_img_path, format='jpeg')
            img = cv2.imread(out_full_img_path)
        return img

    def extract_lines(self, page_id: str, image_data: dict):
        count = 0
        lines = []
        try:
            if self.should_filter_by_class:
                    accepted_zones = []
                    for elt in api_client.paginate('ListElementChildren',id=page_id, with_best_classes=True):
                        printed = True
                        for classification in elt['best_classes']:
                            if classification['ml_class']['name'] == 'handwritten':
                                printed = False
                        for classification in elt['best_classes']:
                            if classification['ml_class']['name'] in self.accepted_classes:
                                if self.should_filter_printed:
                                    if not printed:
                                        accepted_zones.append(elt['zone']['id'])
                                else:
                                    accepted_zones.append(elt['zone']['id'])
                    logger.info('Number of accepted zone for page {} : {}'.format(page_id,len(accepted_zones)))

            for res in api_client.paginate('ListTranscriptions', id=page_id, type='line', recursive=True):
                if self.should_filter_by_slug and res['source']['slug'] not in self.accepted_slugs:
                    continue

                if self.should_filter_by_class and res['zone']['id'] not in accepted_zones:
                    continue
                
                text = res['text']
                if not text or not text.strip():
                    continue

                if res['zone']:
                    polygon = res['zone']['polygon']
                elif res['element']['zone']:
                    polygon = res['element']['zone']['polygon']
                else:
                    raise ValueError(f"Data problem with polygon :: {res}")

                polygon = np.asarray(polygon).clip(0)
                [x, y, w, h] = cv2.boundingRect(polygon)
                lines.append(((x, y, w, h), polygon, text))
                count += 1
        except ErrorResponse as e:
            logger.info(f"ListTranscriptions failed {e.status_code} - {e.title} - {e.content} - {page_id}")
            raise e
        logger.debug(f"Num of lines {count}")
        if count == 0:
            logger.info(f"Page {page_id} skipped, because it has no lines")
            return

        full_image_url = image_data['s3_url']
        if full_image_url is None:
            full_image_url = image_data['url'] + '/full/full/0/default.jpg'

        img = self.get_image(full_image_url, page_id=page_id)

        # sort vertically then horizontally
        sorted_lines = sorted(lines, key=lambda key: (key[0][1], key[0][0]))

        if self.extraction_mode == Extraction.boundingRect:
            for i, ((x, y, w, h), polygon, text) in enumerate(sorted_lines):
                cropped = img[y:y + h, x:x + w].copy()
                cv2.imwrite(f'{self.out_line_img_dir}/{page_id}_{i}.jpg', cropped)

        elif self.extraction_mode == Extraction.polygon:
            for i, (rect, polygon, text) in enumerate(sorted_lines):
                polygon_img = self.extract_polygon_image(img, polygon=polygon, rect=rect)
                cv2.imwrite(f'{self.out_line_img_dir}/{page_id}_{i}.jpg', polygon_img)

        else:
            raise ValueError("Unsupported extraction mode")

        for i, (rect, polygon, text) in enumerate(sorted_lines):
            write_file(f"{self.out_line_text_dir}/{page_id}_{i}.txt", text)

    @staticmethod
    def extract_polygon_image(img: 'np.ndarray', polygon: 'np.ndarray', rect: Box) -> 'np.ndarray':
        pts = polygon.copy()
        [x, y, w, h] = rect
        cropped = img[y:y + h, x:x + w].copy()
        pts = pts - pts.min(axis=0)
        mask = np.zeros(cropped.shape[:2], np.uint8)
        cv2.drawContours(mask, [pts], -1, (255, 255, 255), -1, cv2.LINE_AA)
        dst = cv2.bitwise_and(cropped, cropped, mask=mask)
        bg = np.ones_like(cropped, np.uint8) * 255
        cv2.bitwise_not(bg, bg, mask=mask)
        dst2 = bg + dst
        return dst2

    def run_pages(self, pages: list):
        for page in tqdm.tqdm(pages):
            page_id = page['id']
            image_data = page['zone']['image']
            logger.debug(f"Page {page_id}")
            self.extract_lines(page_id, image_data)

    def run_volumes(self, volume_ids: list):
        for volume_id in tqdm.tqdm(volume_ids):
            logger.info(f"Volume {volume_id}")
            pages = [page for page in api_client.paginate('ListElementChildren', id=volume_id)]
            self.run_pages(pages)

    def run_folders(self, element_ids: list, volume_type: str):
        for elem_id in tqdm.tqdm(element_ids):
            logger.info(f"Folder {elem_id}")
            vol_ids = [page['id'] for page in
                       api_client.paginate('ListElementChildren', id=elem_id, recursive=True, type=volume_type)]
            self.run_volumes(vol_ids)

    def run_corpora(self, corpus_ids: list, volume_type: str):
        for corpus_id in tqdm.tqdm(corpus_ids):
            logger.info(f"Corpus {corpus_id}")
            vol_ids = [vol['id'] for vol in api_client.paginate('ListElements', corpus=corpus_id, type=volume_type)]
            self.run_volumes(vol_ids)


class Split(Enum):
    Train: int = 0
    Test: int = 1
    Validation: int = 2

    @property
    def short_name(self) -> str:
        if self == self.Validation:
            return "val"
        return self.name.lower()


class KaldiPartitionSplitter:

    def __init__(self, out_dir_base='/tmp/kaldi_data', split_train_ratio=0.8, split_test_ratio=0.1, use_existing_split=False):
        self.out_dir_base = out_dir_base
        self.split_train_ratio = split_train_ratio
        self.split_test_ratio = split_test_ratio
        self.split_val_ratio = 1 - self.split_train_ratio - self.split_test_ratio
        self.use_existing_split = use_existing_split

    def page_level_split(self, line_ids: list) -> dict:
        page_ids = list({'_'.join(line_id.split('_')[:-1]) for line_id in line_ids})
        random.shuffle(page_ids)
        page_count = len(page_ids)

        train_page_ids = page_ids[:round(page_count * self.split_train_ratio)]
        page_ids = page_ids[round(page_count * self.split_train_ratio):]

        test_page_ids = page_ids[:round(page_count * self.split_test_ratio)]
        page_ids = page_ids[round(page_count * self.split_test_ratio):]

        val_page_ids = page_ids

        page_dict = {page_id: Split.Train.value for page_id in train_page_ids}
        page_dict.update({page_id: Split.Test.value for page_id in test_page_ids})
        page_dict.update({page_id: Split.Validation.value for page_id in val_page_ids})
        return page_dict

    def existing_split(self, line_ids: list) -> list:
        split_dict = {split.short_name: [] for split in Split}
        for line_id in line_ids:
            split_prefix = line_id.split('/')[0].lower()
            split_dict[split_prefix].append(line_id)
        splits = [split_dict[split.short_name] for split in Split]
        return splits

    def create_partitions(self):
        logger.info("Creating partitions")
        lines_path = Path(f'{self.out_dir_base}/Lines')
        line_ids = [str(file.relative_to(lines_path).with_suffix('')) for file in lines_path.glob('**/*.jpg')]

        if self.use_existing_split:
            logger.info("Using existing split")
            datasets = self.existing_split(line_ids)
        else:
            page_dict = self.page_level_split(line_ids)
            datasets = [[] for _ in range(3)]
            for line_id in line_ids:
                page_id = '_'.join(line_id.split('_')[:-1])
                split_id = page_dict[page_id]
                datasets[split_id].append(line_id)

        partitions_dir = os.path.join(self.out_dir_base, 'Partitions')
        os.makedirs(partitions_dir, exist_ok=True)
        for i, dataset in enumerate(datasets):
            if not dataset:
                logger.info(f"Partition {Split(i).name} is empty! Skipping..")
                continue
            file_name = f"{partitions_dir}/{Split(i).name}Lines.lst"
            write_file(file_name, '\n'.join(dataset) + '\n')


def create_parser():
    parser = argparse.ArgumentParser(
        description="Script to generate Kaldi training data from annotations from Arkindex",
        formatter_class=argparse.ArgumentDefaultsHelpFormatter)
    parser.add_argument('-n', '--dataset_name', type=str, required=True,
                        help='Name of the dataset being created for kaldi '
                             '(useful for distinguishing different datasets when in Lines or Transcriptions directory)')
    parser.add_argument('-o', '--out_dir', type=str, required=True,
                        help='output directory')
    parser.add_argument('--train_ratio', type=float, default=0.8,
                        help='Ratio of pages to be used in train (between 0 and 1)')
    parser.add_argument('--test_ratio', type=float, default=0.1,
                        help='Ratio of pages to be used in test (between 0 and 1 - train_ratio)')
    parser.add_argument('--use_existing_split', action='store_true', default=False,
                        help='Use an existing split instead of random. '
                             'Expecting line_ids to be prefixed with (train, val and test)')
    parser.add_argument('--split_only', '--no_download', action='store_true', default=False,
                        help="Vreate the split from already downloaded lines, don't download the lines")

    parser.add_argument('-e', '--extraction_mode', type=lambda x: Extraction[x], default=Extraction.boundingRect,
                        help=f'Mode for extracting the line images: {[e.name for e in Extraction]}')

    group = parser.add_mutually_exclusive_group(required=False)
    group.add_argument('--grayscale', action='store_true',
                       help='Convert images to grayscale')
    group.add_argument('--color', action='store_false',
                       help='Use color images')
    parser.set_defaults(grayscale=True)

    parser.add_argument('--corpora', nargs='*',
                        help='List of corpus ids to be used, separated by spaces')
    parser.add_argument('--folders', type=str, nargs='*',
                        help='List of folder ids to be used, separated by spaces. '
                             'Elements of `volume_type` will be searched recursively in these folders')
    parser.add_argument('--volumes', nargs='*',
                        help='List of volume ids to be used, separated by spaces')
    # parser.add_argument('--pages', nargs='*',
    #                     help='List of page ids to be used, separated by spaces')
    parser.add_argument('-v', '--volume_type', type=str, default='volume',
                        help='Volumes (1 level above page) may have a different name on corpora')

    parser.add_argument('--accepted_slugs', nargs='*',
                        help='List of accepted slugs for downloading transcriptions')

    parser.add_argument('--accepted_classes', nargs='*',
                        help='List of accepted ml_class names. Filter lines by class of related elements')

    parser.add_argument('--filter_printed', action='store_true',
    help='Filter lines annotated as printed')
    return parser


def main():
    args = create_parser().parse_args()

    logger.info(f"ARGS {args} \n")

    if not args.split_only:
        kaldi_data_generator = KaldiDataGenerator(dataset_name=args.dataset_name,
                                                  out_dir_base=args.out_dir,
                                                  grayscale=args.grayscale,
                                                  extraction=args.extraction_mode,
                                                  accepted_slugs=args.accepted_slugs,
                                                  accepted_classes=args.accepted_classes,
                                                  filter_printed=args.filter_printed)

        # extract all the lines and transcriptions
        # if args.pages:
        #     kaldi_data_generator.run_pages(args.pages)
        if args.volumes:
            kaldi_data_generator.run_volumes(args.volumes)
        if args.folders:
            kaldi_data_generator.run_folders(args.folders, args.volume_type)
        if args.corpora:
            kaldi_data_generator.run_corpora(args.corpora, args.volume_type)
    else:
        logger.info("Creating a split from already downloaded files")

    kaldi_partitioner = KaldiPartitionSplitter(out_dir_base=args.out_dir,
                                               split_train_ratio=args.train_ratio,
                                               split_test_ratio=args.test_ratio,
                                               use_existing_split=args.use_existing_split)

    # create partitions from all the extracted data
    kaldi_partitioner.create_partitions()

    logger.info("DONE")


if __name__ == '__main__':
    main()