-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathunified_layout_evaluation.py
158 lines (135 loc) · 7.24 KB
/
unified_layout_evaluation.py
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
import os
import json
import numpy as np
from tqdm import tqdm
import sys
from detectron2.evaluation import DatasetEvaluator
from detectron2.data import MetadataCatalog
class MyEncoder(json.JSONEncoder):
def default(self, obj):
if isinstance(obj, np.integer):
return int(obj)
elif isinstance(obj, np.floating):
return float(obj)
elif isinstance(obj, np.ndarray):
return obj.tolist()
else:
return super(MyEncoder, self).default(obj)
class UniLayoutEvaluator(DatasetEvaluator):
"""Evaluator for Unified Layout Analysis.
Args:
DatasetEvaluator (_type_): _description_
"""
def __init__(self, dataset_name, output_dir, as_two_stage=False) -> None:
"""
Args:
dataset_name (str): name of the dataset, e.g., "COMP_HRDOC_HR_TEST"
"""
self._dataset_name = dataset_name
self._output_dir = output_dir
self.as_two_stage = as_two_stage
if not os.path.exists(self._output_dir):
os.makedirs(self._output_dir)
def reset(self) -> None:
"""Reset the internal state to prepare for a new round of evaluation.
"""
pass
def process(self, inputs, outputs) -> None:
"""Process an image and its annotations to evaluate.
Args:
inputs (dict): a dict that contains the input image
outputs (dict): a dict that contains the output of the model
"""
if os.path.exists(os.path.join(self._output_dir, 'hr_json')) == False:
os.mkdir(os.path.join(self._output_dir, 'hr_json'))
hr_path = os.path.join(self._output_dir, 'hr_json', inputs[0]['file_name'][0].split('/')[-1].split('_')[0] + ".json")
with open(hr_path, "w") as json_file:
json.dump(outputs['output_for_document'], json_file, indent=4, cls=MyEncoder)
# if self.as_two_stage:
if os.path.exists(os.path.join(self._output_dir, 'det_json')) == False:
os.mkdir(os.path.join(self._output_dir, 'det_json'))
for i, page_level_det in enumerate(outputs['output_for_page']):
if len(page_level_det) == 0:
page_det_path = os.path.join(self._output_dir, 'det_json', outputs['output_for_page'][0][0]['file_name'].split('_')[0] + f'_{i}' + ".json")
else:
page_det_path = os.path.join(self._output_dir, 'det_json', page_level_det[0]['file_name'][:-4] + ".json")
with open(page_det_path, "w") as json_file:
json.dump(page_level_det, json_file, indent=4, cls=MyEncoder)
"""
Table of Contents Extraction
"""
toc_path = os.path.join(self._output_dir, 'toc_json', inputs[0]['file_name'][0].split('/')[-1].split('_')[0] + ".json")
with open(toc_path, "w") as json_file:
json.dump(outputs['output_for_toc'], json_file, indent=4, cls=MyEncoder)
"""
Hierarchical Document Structure Reconstruction
"""
hr_path = os.path.join(self._output_dir, 'hr_json', inputs[0]['file_name'][0].split('/')[-1].split('_')[0] + ".json")
with open(hr_path, "w") as json_file:
json.dump(outputs['output_for_document'], json_file, indent=4, cls=MyEncoder)
return
def evaluate(self) -> None:
"""Evaluate the predictions collected so far.
"""
gt_folder = "datasets/Comp-HRDoc/HRDH_MSRA_POD_TEST/test_eval"
pred_folder = os.path.join(self._output_dir, 'hr_json')
save_folder = os.path.join(self._output_dir, 'logical_role_json')
if os.path.exists(save_folder) == False:
os.mkdir(save_folder)
print('Reading Order Prediction Evaluation on Comp-HRDoc')
print('********************************************************')
cmd = f"{sys.executable} projects/unified_layout_analysis/evaluation/hrdoc_tool/reading_order_eval.py --gt_folder {gt_folder} --pred_folder {pred_folder} --num_workers 8 \n"
os.system(cmd)
if not self.as_two_stage:
gt_jsons = sorted(os.listdir(gt_folder))
for gt_json_file in tqdm(gt_jsons):
gt_json = json.load(open(os.path.join(gt_folder, gt_json_file)))
pred_json_file = gt_json_file
pred_json = json.load(open(os.path.join(pred_folder, pred_json_file)))
rearranged_pred = rearrange_pred(pred_json, gt_json)
with open(os.path.join(save_folder, gt_json_file), 'w') as json_file:
json.dump(rearranged_pred, json_file, indent=4)
print('Logical Role Classification Evaluation on Comp-HRDoc')
print('********************************************************')
cmd = f"{sys.executable} projects/unified_layout_analysis/evaluation/hrdoc_tool/classify_eval.py --gt_folder {gt_folder} --pred_folder {save_folder} \n"
os.system(cmd)
else:
print('Page Object Detection Evaluation on Comp-HRDoc')
print('********************************************************')
pod_gt = "datasets/Comp-HRDoc/HRDH_MSRA_POD_TEST/coco_test.json"
pod_pred = os.path.join(self._output_dir, 'det_json')
cmd = f"{sys.executable} projects/unified_layout_analysis/evaluation/hrdoc_tool/page_object_detection_eval.py --gt_anno {pod_gt} --pred_folder {pod_pred}\n"
os.system(cmd)
print('Table of Contents Extraction Evaluation on Comp-HRDoc')
print('********************************************************')
toc_gt = "datasets/Comp-HRDoc/HRDH_MSRA_POD_TEST/test_eval_toc/"
toc_pred = os.path.join(self._output_dir, 'toc_json')
cmd = f"{sys.executable} projects/unified_layout_analysis/evaluation/hrdoc_tool/teds_eval.py --gt_anno {toc_gt} --pred_folder {toc_pred}\n"
os.system(cmd)
print('Hierarchical Document Structure Reconstruction Evaluation on Comp-HRDoc')
print('********************************************************')
hds_gt = "datasets/Comp-HRDoc/HRDH_MSRA_POD_TEST/test_eval/"
hds_pred = os.path.join(self._output_dir, 'hr_json')
cmd = f"{sys.executable} projects/unified_layout_analysis/evaluation/hrdoc_tool/teds_eval.py --gt_anno {hds_gt} --pred_folder {hds_pred}\n"
return
def rearrange_pred(pred_json, gt_json):
pred_box = [pred_json[i]['box'] for i in range(len(pred_json))]
gt_box = [gt_json[i]['box'] for i in range(len(gt_json))]
assert len(pred_box) == len(gt_box)
rearranged_pred = []
for i, box in enumerate(gt_box):
try:
found = False
for index, p_b in enumerate(pred_box):
if p_b == box and pred_json[index]['page'] == gt_json[i]['page']:
found = True
break
if not found:
print("Box not found in pred_json: {}, Class: {}".format(box, gt_json[i]['class']))
rearranged_pred.append(gt_json[i])
continue
rearranged_pred.append(pred_json[index])
except:
print("Text not found in pred_json: {}".format(box))
rearranged_pred.append(gt_json[i])
return rearranged_pred