-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcompute_metrics.py
262 lines (236 loc) · 16.3 KB
/
compute_metrics.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
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
261
"""
Simple script to compute the metrics on the scores distribution of the detectors.
Authors:
Edoardo Daniele Cannas - edoardodaniele.cannas@polimi.it
"""
# --- Libraries --- #
from utils.metrics import compute_all_metrics
import pandas as pd
import argparse
import sys
import os
import glob
# --- Functions --- #
def main(args: argparse.Namespace):
# --- Parse the arguments --- #
results_dir = args.input_dir
detector = args.detector
debug = args.debug
# --- Load the data --- #
# Pristine ("real") samples
results_pristine = pd.read_csv(os.path.join(results_dir, detector, 'real.csv'), index_col=[0, 1])
results_pristine_jpegai = pd.read_csv(os.path.join(results_dir, detector, 'real_JPEGAI.csv'), index_col=[0, 1])
results_pristine_djpegai = pd.read_csv(os.path.join(results_dir, detector, 'real_doubleJPEGAI.csv'), index_col=[0, 1])
results_pristine_jpeg = pd.read_csv(os.path.join(results_dir, detector, 'real_JPEG.csv'), index_col=[0, 1])
# Synthetic samples
results_synthetic = pd.read_csv(os.path.join(results_dir, detector, 'synthetic.csv'), index_col=[0, 1])
results_synthetic_jpegai = pd.read_csv(os.path.join(results_dir, detector, 'synthetic_JPEGAI.csv'), index_col=[0, 1])
results_synthetic_jpeg = pd.read_csv(os.path.join(results_dir, detector, 'synthetic_JPEG.csv'), index_col=[0, 1])
# --- Compute the metrics --- #
metrics = {}
# --- Considering all images together
dataset_name = 'All_images'
metrics[dataset_name] = {}
metrics[dataset_name]['All'] = []
# Pristine test cases
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_pristine, results_pristine_jpegai)],
index=pd.Index(['Real_vs_Real-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_pristine, results_pristine_djpegai)],
index=pd.Index(['Real_vs_Real-Double-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(pd.DataFrame([compute_all_metrics(results_pristine, results_pristine_jpeg)],
index=pd.Index(['Real_vs_Real-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0',
'fnr_thr0', 'tnr_thr0']))
# Synthetic test cases
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_synthetic, results_synthetic_jpegai)],
index=pd.Index(['Synth_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_synthetic, results_synthetic_jpeg)],
index=pd.Index(['Synth_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Pristine VS Synthetic test cases
metrics[dataset_name]['All'].append(pd.DataFrame([compute_all_metrics(results_pristine, results_synthetic)],
index=pd.Index(['Real_vs_Synth']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0',
'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_pristine_jpeg, results_synthetic_jpeg)],
index=pd.Index(['Real-JPEG_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_pristine_jpegai, results_synthetic_jpegai)],
index=pd.Index(['Real-JPEGAI_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'] = pd.concat(metrics[dataset_name]['All'])
# Divide by quality factor and target_bpp
for idx, target_bpp in enumerate(results_pristine_jpegai['target_bpp'].unique()):
part_results = []
# Pristine test cases
dataset = results_pristine_jpegai.loc[results_pristine_jpegai['target_bpp'] == target_bpp]
part_results.append(pd.DataFrame([compute_all_metrics(results_pristine, dataset)],
index=pd.Index(['Real_vs_Real-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
if target_bpp in results_pristine_djpegai['target_bpp'].unique():
dataset = results_pristine_djpegai.loc[results_pristine_djpegai['target_bpp'] == target_bpp]
part_results.append(pd.DataFrame([compute_all_metrics(results_pristine, dataset)],
index=pd.Index(['Real_vs_Real-Double-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
# Synthetic test cases
dataset = results_synthetic_jpegai.loc[results_synthetic_jpegai['target_bpp'] == target_bpp]
part_results.append(pd.DataFrame([compute_all_metrics(results_synthetic, dataset)],
index=pd.Index(['Synth_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
# Pristine VS Synthetic test cases
dataset_pristine = results_pristine_jpegai.loc[results_pristine_jpegai['target_bpp'] == target_bpp]
dataset_syn = results_synthetic_jpegai.loc[results_synthetic_jpegai['target_bpp'] == target_bpp]
part_results.append(pd.DataFrame([compute_all_metrics(dataset_pristine, dataset_syn)],
index=pd.Index(['Real-JPEGAI_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
metrics[dataset_name][target_bpp] = pd.concat(part_results)
for idx, qf in enumerate(results_pristine_jpeg['qf'].dropna().unique()):
part_results = []
# Pristine test cases
dataset = results_pristine_jpeg.loc[results_pristine_jpeg['qf'] == qf]
part_results.append(pd.DataFrame([compute_all_metrics(results_pristine, dataset)],
index=pd.Index(['Real_vs_Real-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
# Synthetic test cases
dataset = results_synthetic_jpeg.loc[results_synthetic_jpeg['qf'] == qf]
part_results.append(pd.DataFrame([compute_all_metrics(results_synthetic, dataset)],
index=pd.Index(['Synth_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
# Pristine VS Synthetic test cases
dataset_pristine = results_pristine_jpeg.loc[results_pristine_jpeg['qf'] == qf]
dataset_syn = results_synthetic_jpeg.loc[results_synthetic_jpeg['qf'] == qf]
part_results.append(pd.DataFrame([compute_all_metrics(dataset_pristine, dataset_syn)],
index=pd.Index(['Real-JPEG_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0',
'tnr_thr0']))
metrics[dataset_name][qf] = pd.concat(part_results)
# --- Considering the single datasets separately
for dataset_name in results_pristine.index.get_level_values(0).unique():
# Prepare the dict
metrics[dataset_name] = {}
# Consider all the images together
metrics[dataset_name]['All'] = []
# Pristine test cases
metrics[dataset_name]['All'].append(pd.DataFrame(
[compute_all_metrics(results_pristine.loc[dataset_name], results_pristine_jpegai.loc[dataset_name])],
index=pd.Index(['Real_vs_Real-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
if dataset_name in results_pristine_djpegai.index.get_level_values(0).unique():
metrics[dataset_name]['All'].append(pd.DataFrame(
[compute_all_metrics(results_pristine.loc[dataset_name], results_pristine_djpegai.loc[dataset_name])],
index=pd.Index(['Real_vs_Real-Double-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(pd.DataFrame(
[compute_all_metrics(results_pristine.loc[dataset_name], results_pristine_jpeg.loc[dataset_name])],
index=pd.Index(['Real_vs_Real-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Synthetic test cases
metrics[dataset_name]['All'].append(pd.DataFrame(
[compute_all_metrics(results_synthetic.loc[dataset_name], results_synthetic_jpegai.loc[dataset_name])],
index=pd.Index(['Synth_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(pd.DataFrame(
[compute_all_metrics(results_synthetic.loc[dataset_name], results_synthetic_jpeg.loc[dataset_name])],
index=pd.Index(['Synth_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Pristine VS Synthetic test cases
metrics[dataset_name]['All'].append(
pd.DataFrame([compute_all_metrics(results_pristine.loc[dataset_name], results_synthetic.loc[dataset_name])],
index=pd.Index(['Real_vs_Synth']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(pd.DataFrame(
[compute_all_metrics(results_pristine_jpeg.loc[dataset_name], results_synthetic_jpeg.loc[dataset_name])],
index=pd.Index(['Real-JPEG_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name]['All'].append(pd.DataFrame([compute_all_metrics(
results_pristine_jpegai.loc[dataset_name], results_synthetic_jpegai.loc[dataset_name])],
index=pd.Index(['Real-JPEGAI_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Concatenate everything
metrics[dataset_name]['All'] = pd.concat(metrics[dataset_name]['All'])
# Divide by quality factor and target_bpp
for idx, target_bpp in enumerate(results_pristine_jpegai['target_bpp'].unique()):
part_results = []
# Pristine test cases
dataset = results_pristine_jpegai.loc[results_pristine_jpegai['target_bpp'] == target_bpp]
part_results.append(
pd.DataFrame([compute_all_metrics(results_pristine.loc[dataset_name], dataset.loc[dataset_name])],
index=pd.Index(['Real_vs_Real-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
if ((dataset_name in results_pristine_djpegai.index.get_level_values(0).unique()) and
(target_bpp in results_pristine_djpegai['target_bpp'].unique())):
dataset = results_pristine_djpegai.loc[results_pristine_djpegai['target_bpp'] == target_bpp]
part_results.append(
pd.DataFrame([compute_all_metrics(results_pristine.loc[dataset_name], dataset.loc[dataset_name])],
index=pd.Index(['Real_vs_Real-Double-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Synthetic test cases
dataset = results_synthetic_jpegai.loc[results_synthetic_jpegai['target_bpp'] == target_bpp]
part_results.append(
pd.DataFrame([compute_all_metrics(results_synthetic.loc[dataset_name], dataset.loc[dataset_name])],
index=pd.Index(['Synth_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Pristine VS Synthetic test cases
dataset_pristine = results_pristine_jpegai.loc[results_pristine_jpegai['target_bpp'] == target_bpp]
dataset_syn = results_synthetic_jpegai.loc[results_synthetic_jpegai['target_bpp'] == target_bpp]
part_results.append(
pd.DataFrame([compute_all_metrics(dataset_pristine.loc[dataset_name], dataset_syn.loc[dataset_name])],
index=pd.Index(['Real-JPEGAI_vs_Synth-JPEGAI']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name][target_bpp] = pd.concat(part_results)
for idx, qf in enumerate(results_pristine_jpeg['qf'].dropna().unique()):
part_results = []
# Pristine test cases
dataset = results_pristine_jpeg.loc[results_pristine_jpeg['qf'] == qf]
part_results.append(
pd.DataFrame([compute_all_metrics(results_pristine.loc[dataset_name], dataset.loc[dataset_name])],
index=pd.Index(['Real_vs_Real-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Synthetic test cases
dataset = results_synthetic_jpeg.loc[results_synthetic_jpeg['qf'] == qf]
part_results.append(
pd.DataFrame([compute_all_metrics(results_synthetic.loc[dataset_name], dataset.loc[dataset_name])],
index=pd.Index(['Synth_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
# Pristine VS Synthetic test cases
dataset_pristine = results_pristine_jpeg.loc[results_pristine_jpeg['qf'] == qf]
dataset_syn = results_synthetic_jpeg.loc[results_synthetic_jpeg['qf'] == qf]
part_results.append(
pd.DataFrame([compute_all_metrics(dataset_pristine.loc[dataset_name], dataset_syn.loc[dataset_name])],
index=pd.Index(['Real-JPEG_vs_Synth-JPEG']),
columns=['WD', 'auc', 'fpr_thr0', 'tpr_thr0', 'ba_thr0', 'fnr_thr0', 'tnr_thr0']))
metrics[dataset_name][qf] = pd.concat(part_results)
# Create the final results Dataframe
metrics = pd.concat(
[pd.concat({key: pd.concat(metrics[key])}, names=['Dataset', 'Quality', 'Test']) for key in
metrics.keys()])
# --- Save the metrics --- #
metrics_df = pd.concat({detector: metrics}, names=['Detector', 'Dataset', 'Quality', 'Test'])
metrics_df.to_csv(os.path.join(results_dir, detector, 'metrics.csv'))
if debug:
print(metrics_df)
if __name__ == '__main__':
# --- Parse the arguments --- #
parser = argparse.ArgumentParser(description='Compute the metrics on the scores distribution of the detectors.')
parser.add_argument('--input_dir', type=str, help='Directory containing the results of the detectors.',
default='./results')
parser.add_argument('--detector', type=str, help='Name of the detector.', required=True)
parser.add_argument('--debug', action='store_true', help='If set, print the results.')
args = parser.parse_args()
# --- Call the main --- #
main(args)