[Enhancement] Simplify normalized edit distance calculation (#1060)

* simplify normalized edit distance calculation

* update rapidfuzz minimum version
This commit is contained in:
Max Bachmann 2022-06-10 04:16:17 +02:00 committed by GitHub
parent d3f65aaacf
commit 7c5c784a94
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 5 additions and 9 deletions

View File

@ -2,7 +2,7 @@
import re import re
from difflib import SequenceMatcher from difflib import SequenceMatcher
from rapidfuzz import string_metric from rapidfuzz.distance import Levenshtein
from mmocr.utils import is_type_list from mmocr.utils import is_type_list
@ -63,12 +63,8 @@ def count_matches(pred_texts, gt_texts):
match_res['match_word_ignore_case_symbol'] += 1 match_res['match_word_ignore_case_symbol'] += 1
match_res['gt_word_num'] += 1 match_res['gt_word_num'] += 1
# normalized edit distance norm_ed_sum += Levenshtein.normalized_distance(pred_text_lower_ignore,
edit_dist = string_metric.levenshtein(pred_text_lower_ignore, gt_text_lower_ignore)
gt_text_lower_ignore)
norm_ed = float(edit_dist) / max(1, len(gt_text_lower_ignore),
len(pred_text_lower_ignore))
norm_ed_sum += norm_ed
# number to calculate char level recall & precision # number to calculate char level recall & precision
match_res['gt_char_num'] += len(gt_text_lower_ignore) match_res['gt_char_num'] += len(gt_text_lower_ignore)

View File

@ -6,7 +6,7 @@ matplotlib
mmcv mmcv
mmdet mmdet
pyclipper pyclipper
rapidfuzz rapidfuzz>=2.0.0
regex regex
scikit-image scikit-image
scipy scipy

View File

@ -6,5 +6,5 @@ numpy
opencv-python != 4.5.5.* opencv-python != 4.5.5.*
pyclipper pyclipper
pycocotools pycocotools
rapidfuzz rapidfuzz>=2.0.0
scikit-image scikit-image