205 lines
7.6 KiB
Python
205 lines
7.6 KiB
Python
# https://stackoverflow.com/questions/774316/python-difflib-highlighting-differences-inline
|
|
import difflib
|
|
import unidecode
|
|
import os
|
|
import glob
|
|
import pandas as pd
|
|
import logging
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
VOWELS = 'aeouiy' + 'AEOUIY'
|
|
CONSONANTS = 'bcdfghjklmnpqrstvxwz' + 'BCDFGHJKLMNPQRSTVXWZ'
|
|
# PREDICT_PATH = 'ocr/result'
|
|
# GROUNDTRUTH_PATH = '/mnt/hdd2T/AICR/Datasets/wiki/ground_truth'
|
|
PREDICT_PATH = 'ocr/result/cinamon'
|
|
GROUNDTRUTH_PATH = '/mnt/hdd2T/AICR/Datasets/Backup/1.Hand_writing/Lines/cinnamon_data'
|
|
# note that we also use different preprocess for cinamon data
|
|
# SAVE_PATH = 'wiki_diff'
|
|
SAVE_PATH = 'wiki_diff/cinamon'
|
|
RES_PATH = f'{SAVE_PATH}/result/'
|
|
WRONG_ACCENT_FILE = f'{SAVE_PATH}/wrong_accent.txt'
|
|
LOST_ACCENT_FILE = f'{SAVE_PATH}/lost_accent.txt'
|
|
TOTAL_WORD = 0
|
|
|
|
|
|
def write_accent_error(path, err):
|
|
# path should be wrong_accent_file or lost_accent_file
|
|
with open(path, 'a') as f:
|
|
f.write(err)
|
|
f.write('\n')
|
|
|
|
|
|
def update_ddata_specialchars(ddata_specialchars, correction_key, char_key):
|
|
if char_key in ddata_specialchars[correction_key]:
|
|
ddata_specialchars[correction_key][char_key] += 1
|
|
else:
|
|
ddata_specialchars[correction_key][char_key] = 1
|
|
|
|
|
|
def process_replace_tag(matcher, i1, i2, j1, j2, ddata, ddata_specialchars):
|
|
a_char = matcher.a[i1:i2]
|
|
b_char = matcher.b[j1:j2]
|
|
ddata['res_text'] += ' ### {' + a_char + ' -> ' + b_char + '} ### '
|
|
ddata['nwrongs'] += 1*len(b_char)
|
|
if len(a_char) == 1 and len(b_char) == 1: # single char case
|
|
if a_char.lower() == b_char.lower(): # wrong upper/lower case
|
|
ddata['UL_single'] += 1
|
|
update_ddata_specialchars(ddata_specialchars, 'UL', (a_char, b_char))
|
|
else:
|
|
ddata['nwrongs_single'] += 1
|
|
a_ori = unidecode.unidecode(a_char).lower()
|
|
b_ori = unidecode.unidecode(b_char).lower()
|
|
if a_ori in VOWELS and b_ori in VOWELS:
|
|
if a_ori == b_ori:
|
|
err = a_char + ' -> ' + b_char
|
|
if b_ori == b_char.lower(): # e.g. Ơ -> O
|
|
ddata['nlost_accent'] += 1
|
|
# write_accent_error(LOST_ACCENT_FILE, err)
|
|
else: # e.g Ơ -> Ớ
|
|
ddata['nwrong_accent'] += 1
|
|
# write_accent_error(WRONG_ACCENT_FILE, err)
|
|
else: # e.g Ă -> Â
|
|
ddata['nwrong_vowels'] += 1
|
|
else:
|
|
if a_ori in CONSONANTS and b_ori in CONSONANTS:
|
|
ddata['nwrong_consonants'] += 1
|
|
else:
|
|
ddata['nwrong_specialchars'] += 1
|
|
update_ddata_specialchars(ddata_specialchars, 'wrong', (a_char, b_char))
|
|
else:
|
|
if a_char.lower() == b_char.lower():
|
|
ddata['UL_multiple'] += 1
|
|
update_ddata_specialchars(ddata_specialchars, 'UL', (a_char, b_char))
|
|
else:
|
|
ddata['nwrongs_multiple'] += 1
|
|
if len(a_char) > 10 or len(b_char) > 10:
|
|
ddata['nlong_sequences'] += 1
|
|
# print(a_char)
|
|
|
|
|
|
def process_delete_tag(matcher, i1, i2, ddata, ddata_specialchars):
|
|
a_char = matcher.a[i1:i2]
|
|
ddata['res_text'] += ' ### {- ' + a_char + '} ### '
|
|
ddata['nadds'] += 1*len(a_char)
|
|
if len(a_char) == 1:
|
|
ddata['nadds_single'] += 1
|
|
if a_char.lower() in CONSONANTS + VOWELS:
|
|
ddata['nadds_chars'] += 1
|
|
else:
|
|
if a_char == ' ':
|
|
ddata['nadds_space'] += 1
|
|
else:
|
|
ddata['nadds_specialchars'] += 1
|
|
update_ddata_specialchars(ddata_specialchars, 'add', a_char)
|
|
|
|
else:
|
|
ddata['nadds_multiple'] += 1
|
|
if len(a_char) > 10:
|
|
ddata['nlong_sequences'] += 1
|
|
# print(a_char)
|
|
|
|
|
|
def process_insert_tag(matcher, j1, j2, ddata, ddata_specialchars):
|
|
b_char = matcher.b[j1:j2]
|
|
ddata['nlosts'] += 1*len(b_char)
|
|
ddata['res_text'] += ' ### {+ ' + b_char + '} ### '
|
|
if len(b_char) == 1:
|
|
ddata['nlosts_single'] += 1
|
|
if b_char.lower() in CONSONANTS + VOWELS:
|
|
ddata['nlosts_chars'] += 1
|
|
else:
|
|
if b_char == ' ':
|
|
ddata['nlosts_space'] += 1
|
|
else:
|
|
ddata['nlosts_specialchars'] += 1
|
|
update_ddata_specialchars(ddata_specialchars, 'lost', b_char)
|
|
|
|
else:
|
|
ddata['nlosts_multiple'] += 1
|
|
if len(b_char) > 10:
|
|
ddata['nlong_sequences'] += 1
|
|
# print(b_char)
|
|
|
|
|
|
def inline_diff(a, b, ddata_specialchars={'lost': {}, 'add': {}, 'wrong': {}, 'UL': {}}):
|
|
matcher = difflib.SequenceMatcher(None, a, b)
|
|
ddata = {'res_text': ''}
|
|
# ddata = ddata | {key: 0 for key in ['nsingle', 'nmultiple']}
|
|
ddata = ddata | {key: 0 for key in ['UL_single', 'UL_multiple']}
|
|
ddata = ddata | {
|
|
key: 0 for key in
|
|
['nlosts', 'nlosts_single', 'nlosts_multiple', 'nlosts_chars', 'nlosts_specialchars', 'nlosts_space']}
|
|
ddata = ddata | {
|
|
key: 0 for key in
|
|
['nadds', 'nadds_single', 'nadds_multiple', 'nadds_chars', 'nadds_specialchars', 'nadds_space']}
|
|
ddata = ddata | {
|
|
key: 0 for key in
|
|
['nwrongs', 'nwrongs_single', 'nwrongs_multiple', 'nwrong_accent', 'nlost_accent', 'nwrong_vowels',
|
|
'nwrong_consonants', 'nwrong_specialchars']}
|
|
ddata['nlong_sequences'] = 0
|
|
for tag, i1, i2, j1, j2 in matcher.get_opcodes():
|
|
if tag == 'replace': # wrong
|
|
process_replace_tag(matcher, i1, i2, j1, j2, ddata, ddata_specialchars)
|
|
if tag == 'delete': # OCR add char so the matcher "delete"
|
|
process_delete_tag(matcher, i1, i2, ddata, ddata_specialchars)
|
|
if tag == 'equal':
|
|
ddata['res_text'] += matcher.a[i1:i2]
|
|
if tag == 'insert': # OCR lost char so the matcher "insert"
|
|
process_insert_tag(matcher, j1, j2, ddata, ddata_specialchars)
|
|
ddata["ned"] = ddata['nwrongs'] + ddata['nadds'] + ddata['nlosts']
|
|
return ddata
|
|
|
|
|
|
def process_single_file(file_name, ddata_specialchars):
|
|
|
|
# read predict file
|
|
with open(os.path.join(PREDICT_PATH, file_name), 'r') as f:
|
|
predict = f.readlines()[0].strip()
|
|
# predict = ''.join(predict)
|
|
# predict = predict.replace(' ', '')
|
|
# predict = predict.replace('\n', '')
|
|
# print(predict)
|
|
|
|
# read groundtruth file
|
|
with open(os.path.join(GROUNDTRUTH_PATH, file_name), 'r') as f:
|
|
gt = f.readlines()[0].strip()
|
|
# gt = ''.join(gt)
|
|
# gt = gt.replace('\n', '')
|
|
|
|
# get statiscal data of difference between predict and ground truth
|
|
ddata = inline_diff(predict, gt, ddata_specialchars)
|
|
global TOTAL_WORD
|
|
TOTAL_WORD = TOTAL_WORD + len(gt.split())
|
|
# write to save_path
|
|
res_text = ddata.pop('res_text', None)
|
|
save_file = os.path.join(RES_PATH, file_name)
|
|
with open(save_file, 'w') as f:
|
|
f.write(res_text)
|
|
|
|
# generate csv file
|
|
ddata = {'file_name': save_file} | ddata
|
|
return ddata
|
|
|
|
|
|
def main(overwrite=False):
|
|
for accent_file in [WRONG_ACCENT_FILE, LOST_ACCENT_FILE]:
|
|
if os.path.exists(accent_file):
|
|
os.remove(accent_file)
|
|
lddata = []
|
|
ddata_specialchars = {'lost': {}, 'add': {}, 'wrong': {}, 'UL': {}}
|
|
for file_ in glob.glob(f'{PREDICT_PATH}/*.txt'):
|
|
file_name = file_.split('/')[-1]
|
|
ddata = process_single_file(file_name, ddata_specialchars)
|
|
lddata.append(ddata)
|
|
if overwrite:
|
|
df = pd.DataFrame(lddata)
|
|
df.to_csv(f'{SAVE_PATH}/wiki_diff.csv', sep='\t')
|
|
df_ = pd.DataFrame(ddata_specialchars)
|
|
df_.to_csv(f'{SAVE_PATH}/wiki_diff_specialchars.csv', sep='\t')
|
|
logger.info(TOTAL_WORD)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main(overwrite=True)
|