chickenflyshigh/onscreen-translator
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

IMAGE_CHANGE_THRESHOLD exception handled + japanese romanisation fixed and unicode range updated

Browse Source
This commit is contained in:
chickenflyshigh
2024-11-08 13:20:03 +11:00
parent 7e80713191
commit ecc264cf65
9 changed files with 45 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
helpers/batching.py
+3 -2
View File
@@ -195,11 +195,12 @@ Expected format:
["translation1", "translation2", ...]
Translation:"""
response = await self._request(prompt)
try:
response = await self._request(prompt)
response_list = ast.literal_eval(response.strip())
except Exception as e:
logger.error(f"Failed to evaluate response from {self.model} from {self.site}. Error: {e}. Response: {response}")
logger.error(f"Failed to evaluate response from {self.model} from {self.site}. Error: {e}.")
return (1, [], 99999)
logger.debug(repr(self))
logger.info(f'{self.model} translated texts from: {texts_to_translate} to {response_list}.')
helpers/ocr.py
+5 -7
View File
@@ -34,7 +34,8 @@ def _easy_init(easy_languages: list, use_GPU=True, **kwargs):
return easyocr.Reader(easy_languages, gpu=use_GPU, **kwargs)
def _easy_ocr(ocr,image) -> list:
return ocr.readtext(image)
detected_texts = ocr.readtext(image)
return detected_texts
# RapidOCR mostly for mandarin and some other asian languages
# default only supports chinese and english
@@ -116,9 +117,6 @@ def get_confidences(ocr_output) -> list:
if __name__ == '__main__':
# OCR_languages = ['ch_sim','en']
# image_old = '/home/James/Pictures/Screenshots/DP-1.jpg'
# reader = easyocr.Reader(OCR_languages, gpu=True) # this needs to run only once to load the model into memory
# result = reader.readtext(image_old)
# print(result)
print(id_keep_source_lang(init_OCR(model='paddle', paddle_lang='zh', easy_languages=['en', 'ch_sim']), '/home/James/Pictures/Screenshots/DP-1.jpg', 'ch_sim'))
OCR_languages = ['ch_sim','en']
reader = easyocr.Reader(OCR_languages, gpu=True)
helpers/translation.py
+3 -2
View File
@@ -71,6 +71,7 @@ async def translate_API_LLM(texts_to_translate: List[str],
result = await task
logger.debug(f'Result: {result}')
if result is not None:
tasks.discard(task)
translation_attempts += 1
status_code, translations, translation_mismatches = result
if status_code == 0:
@@ -79,14 +80,14 @@ async def translate_API_LLM(texts_to_translate: List[str],
t.cancel()
return translations
else:
logger.error(f"Model has failed to translate the text.")
logger.error(f"Model has failed to translate the text. Result: {result}")
if translation_attempts == no_of_models:
if best_translation is not None:
return translations
else:
logger.error("All models have failed to translate the text.")
raise TypeError("Models have likely all outputted garbage translations or rate limited.")
elif status_code == 1:
elif status_code == 2:
if best_translation is None:
best_translation = (translations, translation_mismatches)
else:
helpers/utils.py
+12 -9
View File
@@ -28,7 +28,7 @@ def is_wayland():
# please install grim otherwise this is way too slow for wayland
def printsc_wayland(region: tuple, path: str):
subprocess.run(['grim','-g', f'{region[0]},{region[1]} {region[2]-region[0]}x{region[3]-region[1]}', '-t', 'jpeg', '-q','95', path])
subprocess.run(['grim','-g', f'{region[0]},{region[1]} {region[2]-region[0]}x{region[3]-region[1]}', '-t', 'jpeg', '-q','100', path])
def printsc_non_wayland(region: tuple, path: str):
# use mss to capture the screen
@@ -94,7 +94,7 @@ def romanize(text, lang):
if lang in ['zh','ch_sim','ch_tra']:
return ' '.join([ py[0] for py in pinyin(text, heteronym=True)])
if lang == 'ja':
return kks.convert(text)[0]['hepburn']
return ' '.join([romaji['hepburn'] for romaji in kks.convert(text)])
return uroman.romanize_string(text)
# check if a string contains words from a language
@@ -103,7 +103,7 @@ def contains_lang(text, lang):
if lang == 'zh':
return bool(re.search(r'[\u4e00-\u9fff]', text))
elif lang == 'ja':
return bool(re.search(r'[\u3040-\u30ff]', text))
return bool(re.search(r'[\u3040-\u30ff]', text)) or bool(re.search(r'[\u4e00-\u9fff]', text))
elif lang == 'ko':
return bool(re.search(r'[\uac00-\ud7af]', text))
elif lang == 'en':
@@ -132,7 +132,7 @@ def standardize_lang(lang):
translation_model_lang = 'ja'
id_model_lang = 'ja'
elif lang == 'ko':
easyocr_lang = 'korean'
easyocr_lang = 'ko'
paddleocr_lang = 'korean'
rapidocr_lang = 'ko'
translation_model_lang = 'ko'
@@ -161,7 +161,7 @@ def which_ocr_lang(model):
else:
raise ValueError("Invalid OCR model. Please use one of 'easy', 'paddle', or 'rapid'.")
def similar_tfidf(list1,list2,threshold) -> float:
def similar_tfidf(list1,list2) -> float:
"""Calculate cosine similarity using TF-IDF vectors."""
if not list1 or not list2:
return 0.0
@@ -174,7 +174,7 @@ def similar_tfidf(list1,list2,threshold) -> float:
vec1 = np.mean(tfidf_matrix[:len(list1)].toarray(), axis=0).reshape(1, -1)
vec2 = np.mean(tfidf_matrix[len(list1):].toarray(), axis=0).reshape(1, -1)
return float(cosine_similarity(vec1, vec2)[0, 0]) > threshold
return cosine_similarity(vec1, vec2)[0, 0]
def similar_jacard(list1, list2) -> float:
if not list1 or not list2:
@@ -183,7 +183,11 @@ def similar_jacard(list1, list2) -> float:
def check_similarity(list1, list2, threshold, method = 'tfidf'):
if method == 'tfidf':
return True if similar_tfidf(list1, list2) >= threshold else False
try:
confidence = similar_tfidf(list1, list2)
except ValueError:
return True
return True if confidence > threshold else False
elif method == 'jacard':
return True if similar_jacard(list1, list2) >= threshold else False
else:
@@ -191,5 +195,4 @@ def check_similarity(list1, list2, threshold, method = 'tfidf'):
if __name__ == "__main__":
# Example usage
x = printsc_wayland((0,0,1920,1080), save = False)
print(x)
print(romanize(lang='ja', text='世界はひろい'))