Update model to use PaddleOCR results

This commit is contained in:
Yun 2021-07-16 16:58:44 +02:00
parent b5e6f5a57f
commit 9b37319961
1 changed files with 5 additions and 15 deletions

View File

@ -17,25 +17,15 @@ class PredictedFrame:
confidence: int # total confidence of all words
text: str
def __init__(self, index: int, pred_data: str, conf_threshold: int):
def __init__(self, index: int, pred_data: list[list], conf_threshold: int):
self.index = index
self.words = []
block = 0 # keep track of line breaks
for l in pred_data.splitlines()[1:]:
word_data = l.split()
if len(word_data) < 12:
# no word is predicted
for l in pred_data:
if len(l) < 2:
continue
_, _, block_num, *_, conf, text = word_data
block_num, conf = int(block_num), int(conf)
# handle line breaks
if block < block_num:
block = block_num
if self.words and self.words[-1].text != '\n':
self.words.append(PredictedWord(0, '\n'))
text = l[1][0]
conf = int(l[1][1] * 100)
# word predictions with low confidence will be filtered out
if conf >= conf_threshold: