def eval_coco_online():
from googletrans import Translator
from time import sleep
# Read in the file
file = open(
'/home/jxgu/github/im2text_jxgu/pytorch/tmp/coco_test_5k_zh.txt', 'r')
zh_lines = file.readlines()
en_lines = []
count = 0
translator = Translator()
for line in zh_lines:
translation = translator.translate(line)
#en_lines.append(translation.text.replace("there is", ""))
en_lines.append(translation.text)
count += 1
if count % 100 == 0:
translator = Translator()
print('... %d sentences prepared' % count)
tmp_name = 'tmp/coco_test_5k_en_online'
with open(tmp_name + '.txt', 'w') as file:
for line in en_lines:
file.write("%s\n" % line.encode("utf-8").lower())
ref_id_json = 'tmp/captions_val_image_info_karpathy_5k_test_11080899_results.json'
utils.text2textid(tmp_name, ref_id_json)
print('Calculating scores for the generated results ... ...')
utils.test2cocojson(tmp_name + '_id')
lang_stats = language_eval_json(tmp_name)
def eval_coco_online(mscoco_zh_text):
from googletrans import Translator
from time import sleep
# Read in the file
file = open(mscoco_zh_text, 'r')
zh_lines = file.readlines()
en_lines = []
count = 0
translator = Translator()
for line in zh_lines:
translation = translator.translate(line)
#en_lines.append(translation.text.replace("there is", ""))
en_lines.append(translation.text)
count += 1
if count % 100 == 0:
translator = Translator()
print('... %d sentences prepared' % count)
tmp_name = mscoco_zh_text.replace('zh', 'en').replace('.txt', 'offline')
with open(tmp_name + '.txt', 'w') as file:
for line in en_lines:
file.write("%s\n" % line.encode("utf-8").lower())
print('Calculating scores for the generated results ... ...')
utils.text2textid(tmp_name, mscoco_zh_text.replace('.txt', '.json'))
utils.test2cocojson(tmp_name + '_id')
lang_stats = language_eval('coco', json.load(open(tmp_name + '_id.json')),
opt.id, '')
def eval_coco_offline():
#os.system("bash test.sh")
from googletrans import Translator
translator = Translator()
root = os.getcwd() + '/'
use_translation = False
mscoco_src_text = '/home/jxgu/github/im2text_jxgu/pytorch/tmp/coco_test_5k_zh.txt'
if os.path.exists(mscoco_src_text) is True and use_translation:
#mscoco_zh_json = 'tmp/20180419-075726.denseatt_zh_mscoco.json'
#utils.cocojson2text(mscoco_zh_json, mscoco_zh_json.replace('.json', '.txt'))
nmt_model = root + "neural_machine_translation/save/20180308-091231/demo-model-0303-full_acc_54.75_ppl_9.10_e22.pt"
print("Start translating chinese to english ...")
bashCommand = "cd neural_machine_translation && python translate.py" + \
" -model " + nmt_model + \
" -src " + mscoco_src_text + \
" -output " + root + 'tmp/coco_test_5k_en.txt' \
" -verbose -gpu 0"
_output = subprocess.check_output(['bash', '-c', bashCommand])
print("Finish translating chinese to english ...")
self_bleu_flag = False
en_lines = []
print('------------------------------------------- aic self-bleu')
if self_bleu_flag:
file = open('tmp/coco_test_5k_zh_en.txt', 'r')
zh_lines = file.readlines()
count = 0
for line in zh_lines:
translation = line
en_lines.append(translation.replace("there is", ""))
count += 1
if count % 100 == 0:
print('... %d sentences prepared' % count)
reference = list()
for line in en_lines:
text = nltk.word_tokenize(line)
reference.append(text)
for i in range(7):
print('Self BLEU: {}'.format(utils.self_bleu(reference, i)))
else:
tmp_name = 'tmp/coco_test_5k_en_offline'
file = open('tmp/coco_test_5k_zh.en.txt', 'r')
en_lines = file.readlines()
with open(tmp_name + '.txt', 'w') as file:
for line in en_lines:
file.write("%s" % line.encode("utf-8").lower())
ref_id_json = 'tmp/captions_val_image_info_karpathy_5k_test_11080899_results.json'
utils.text2textid(tmp_name, ref_id_json)
print('Calculating scores for the generated results ... ...')
utils.test2cocojson(tmp_name + '_id')
lang_stats = language_eval_json(tmp_name)
print('------------------------------------------- end')
def eval_coco_offline():
dump_json_path = i2t_eval()
root = os.getcwd() + '/'
use_translation = False
mscoco_zh_json = dump_json_path
mscoco_zh_text = dump_json_path.replace('.json', '.txt')
utils.cocojson2text(mscoco_zh_json,
mscoco_zh_json.replace('.json', '.txt'))
nmt_model = root + "save/opennmt/aic_zh2en_nmt_part_acc_61.49_ppl_7.43_e22.pt"
print("Start translating chinese to english ...")
bashCommand = "cd ../OpenNMT-py && python translate.py" + \
" -model " + nmt_model + \
" -src " + root + '/' + mscoco_zh_text + \
" -output " + root + '/' + mscoco_zh_text.replace('zh', 'en') + \
" -verbose -gpu 0"
_output = subprocess.check_output(['bash', '-c', bashCommand])
print("Finish translating chinese to english ...")
self_bleu_flag = False
en_lines = []
if self_bleu_flag:
print('------------------------------------------- aic self-bleu')
count = 0
file = open(mscoco_zh_text.replace('zh', 'en'), 'r')
zh_lines = file.readlines()
for line in zh_lines:
translation = line
en_lines.append(translation.replace("there is", ""))
count += 1
if count % 100 == 0:
print('... %d sentences prepared' % count)
reference = list()
for line in en_lines:
text = nltk.word_tokenize(line)
reference.append(text)
for i in range(7):
print('Self BLEU: {}'.format(utils.self_bleu(reference, i)))
else:
tmp_name = mscoco_zh_text.replace('zh',
'en').replace('.txt', 'offline')
file = open(mscoco_zh_text.replace('zh', 'en'), 'r')
en_lines = file.readlines()
with open(tmp_name + '.txt', 'w') as file:
for line in en_lines:
file.write("%s" % line.encode("utf-8").lower())
utils.text2textid(tmp_name, dump_json_path)
print('Calculating scores for the generated results ... ...')
utils.test2cocojson(tmp_name + '_id')
lang_stats = language_eval('coco',
json.load(open(tmp_name + '_id.json')),
opt.id, '')
def eval_30K(type, text_in):
import time
import re, string
if type == 'online':
from googletrans import Translator
translator = Translator()
# Read in the file
file = open(text_in, 'r')
zh_lines = file.readlines()
en_lines = []
count = 0
for line in zh_lines:
if type == 'online':
translation = translator.translate(line)
#en_lines.append(translation.text.replace("there is", ""))
tmp = translation.text
#tmp = re.sub('[%s]' % re.escape(string.punctuation), '', tmp)
else:
tmp = line.replace("there is", "")
en_lines.append(tmp)
count += 1
if count % 100 == 0:
if type == 'online':
translator = Translator()
print('... %d sentences prepared' % count)
tmp_name = 'tmp/flickr_test_1k_en_' + type
with open(tmp_name + '.txt', 'w') as file:
for line in en_lines:
file.write("%s\n" % line.encode("utf-8").lower())
ref_id_json = 'tmp/captions_val_image_info_karpathy_1k_test_11080899_results.json'
utils.text2textid(tmp_name, ref_id_json)
print('Calculating scores for the generated results ... ...')
utils.test2cocojson(tmp_name + '_id')
lang_stats = language_eval_json_30K(tmp_name)