File size: 8,259 Bytes
a89d9fd |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 |
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import re
import sys
import shapely
from shapely.geometry import Polygon
import numpy as np
from collections import defaultdict
import operator
from rapidfuzz.distance import Levenshtein
import argparse
import json
import copy
def parse_ser_results_fp(fp, fp_type="gt", ignore_background=True):
# img/zh_val_0.jpg {
# "height": 3508,
# "width": 2480,
# "ocr_info": [
# {"text": "Maribyrnong", "label": "other", "bbox": [1958, 144, 2184, 198]},
# {"text": "CITYCOUNCIL", "label": "other", "bbox": [2052, 183, 2171, 214]},
# ]
assert fp_type in ["gt", "pred"]
key = "label" if fp_type == "gt" else "pred"
res_dict = dict()
with open(fp, "r", encoding='utf-8') as fin:
lines = fin.readlines()
for _, line in enumerate(lines):
img_path, info = line.strip().split("\t")
# get key
image_name = os.path.basename(img_path)
res_dict[image_name] = []
# get infos
json_info = json.loads(info)
for single_ocr_info in json_info["ocr_info"]:
label = single_ocr_info[key].upper()
if label in ["O", "OTHERS", "OTHER"]:
label = "O"
if ignore_background and label == "O":
continue
single_ocr_info["label"] = label
res_dict[image_name].append(copy.deepcopy(single_ocr_info))
return res_dict
def polygon_from_str(polygon_points):
"""
Create a shapely polygon object from gt or dt line.
"""
polygon_points = np.array(polygon_points).reshape(4, 2)
polygon = Polygon(polygon_points).convex_hull
return polygon
def polygon_iou(poly1, poly2):
"""
Intersection over union between two shapely polygons.
"""
if not poly1.intersects(
poly2): # this test is fast and can accelerate calculation
iou = 0
else:
try:
inter_area = poly1.intersection(poly2).area
union_area = poly1.area + poly2.area - inter_area
iou = float(inter_area) / union_area
except shapely.geos.TopologicalError:
# except Exception as e:
# print(e)
print('shapely.geos.TopologicalError occurred, iou set to 0')
iou = 0
return iou
def ed(args, str1, str2):
if args.ignore_space:
str1 = str1.replace(" ", "")
str2 = str2.replace(" ", "")
if args.ignore_case:
str1 = str1.lower()
str2 = str2.lower()
return Levenshtein.distance(str1, str2)
def convert_bbox_to_polygon(bbox):
"""
bbox : [x1, y1, x2, y2]
output: [[x1, y1], [x2, y2], [x3, y3], [x4, y4]]
"""
xmin, ymin, xmax, ymax = bbox
poly = [[xmin, ymin], [xmax, ymin], [xmax, ymax], [xmin, ymax]]
return poly
def eval_e2e(args):
# gt
gt_results = parse_ser_results_fp(args.gt_json_path, "gt",
args.ignore_background)
# pred
dt_results = parse_ser_results_fp(args.pred_json_path, "pred",
args.ignore_background)
iou_thresh = args.iou_thres
num_gt_chars = 0
gt_count = 0
dt_count = 0
hit = 0
ed_sum = 0
for img_name in dt_results:
gt_info = gt_results[img_name]
gt_count += len(gt_info)
dt_info = dt_results[img_name]
dt_count += len(dt_info)
dt_match = [False] * len(dt_info)
gt_match = [False] * len(gt_info)
all_ious = defaultdict(tuple)
# gt: {text, label, bbox or poly}
for index_gt, gt in enumerate(gt_info):
if "poly" not in gt:
gt["poly"] = convert_bbox_to_polygon(gt["bbox"])
gt_poly = polygon_from_str(gt["poly"])
for index_dt, dt in enumerate(dt_info):
if "poly" not in dt:
dt["poly"] = convert_bbox_to_polygon(dt["bbox"])
dt_poly = polygon_from_str(dt["poly"])
iou = polygon_iou(dt_poly, gt_poly)
if iou >= iou_thresh:
all_ious[(index_gt, index_dt)] = iou
sorted_ious = sorted(
all_ious.items(), key=operator.itemgetter(1), reverse=True)
sorted_gt_dt_pairs = [item[0] for item in sorted_ious]
# matched gt and dt
for gt_dt_pair in sorted_gt_dt_pairs:
index_gt, index_dt = gt_dt_pair
if gt_match[index_gt] == False and dt_match[index_dt] == False:
gt_match[index_gt] = True
dt_match[index_dt] = True
# ocr rec results
gt_text = gt_info[index_gt]["text"]
dt_text = dt_info[index_dt]["text"]
# ser results
gt_label = gt_info[index_gt]["label"]
dt_label = dt_info[index_dt]["pred"]
if True: # ignore_masks[index_gt] == '0':
ed_sum += ed(args, gt_text, dt_text)
num_gt_chars += len(gt_text)
if gt_text == dt_text:
if args.ignore_ser_prediction or gt_label == dt_label:
hit += 1
# unmatched dt
for tindex, dt_match_flag in enumerate(dt_match):
if dt_match_flag == False:
dt_text = dt_info[tindex]["text"]
gt_text = ""
ed_sum += ed(args, dt_text, gt_text)
# unmatched gt
for tindex, gt_match_flag in enumerate(gt_match):
if gt_match_flag == False:
dt_text = ""
gt_text = gt_info[tindex]["text"]
ed_sum += ed(args, gt_text, dt_text)
num_gt_chars += len(gt_text)
eps = 1e-9
print("config: ", args)
print('hit, dt_count, gt_count', hit, dt_count, gt_count)
precision = hit / (dt_count + eps)
recall = hit / (gt_count + eps)
fmeasure = 2.0 * precision * recall / (precision + recall + eps)
avg_edit_dist_img = ed_sum / len(gt_results)
avg_edit_dist_field = ed_sum / (gt_count + eps)
character_acc = 1 - ed_sum / (num_gt_chars + eps)
print('character_acc: %.2f' % (character_acc * 100) + "%")
print('avg_edit_dist_field: %.2f' % (avg_edit_dist_field))
print('avg_edit_dist_img: %.2f' % (avg_edit_dist_img))
print('precision: %.2f' % (precision * 100) + "%")
print('recall: %.2f' % (recall * 100) + "%")
print('fmeasure: %.2f' % (fmeasure * 100) + "%")
return
def parse_args():
"""
"""
def str2bool(v):
return v.lower() in ("true", "t", "1")
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument(
"--gt_json_path",
default=None,
type=str,
required=True, )
parser.add_argument(
"--pred_json_path",
default=None,
type=str,
required=True, )
parser.add_argument("--iou_thres", default=0.5, type=float)
parser.add_argument(
"--ignore_case",
default=False,
type=str2bool,
help="whether to do lower case for the strs")
parser.add_argument(
"--ignore_space",
default=True,
type=str2bool,
help="whether to ignore space")
parser.add_argument(
"--ignore_background",
default=True,
type=str2bool,
help="whether to ignore other label")
parser.add_argument(
"--ignore_ser_prediction",
default=False,
type=str2bool,
help="whether to ignore ocr pred results")
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parse_args()
eval_e2e(args)
|