-
Notifications
You must be signed in to change notification settings - Fork 1
/
Write_results_csv.py
272 lines (206 loc) · 9.28 KB
/
Write_results_csv.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
262
263
264
265
266
267
268
269
270
271
272
"""
Write_results_csv.py
Creates results dataframe from directories and format for ground truth boxes and predicted boxes.
done in Evaluation notebook.
Charlotte Weil, August 2019
- - - - -
Args:
path_to_root = '../../..'
GT_bbox_Dir = os.path.join(path_to_root, 'data/yolov3-inputs_imagery-7-25_cropped_419/validation_set/labels/')
Pred_bbox_Dir = os.path.join(path_to_root, 'outputs/fasterRCNN_07_27_newimagery/export_inference/validationset_inference_fasterRCNN_07_27_newimagery')
gt_format = 'xywh_pix'
pred_format = 'y1x1y2x2_pix'
outputFile = os.path.join(path_to_root, 'results/results_validation_fasterRCNN-07-27_.csv')
detect_df = make_results_table(GT_bbox_Dir, Pred_bbox_Dir, outputFile, gt_format, pred_format)
Outputs:
a cresults_table.csv with columns:
cols = ['img_id','gt_bbox','gt_format','pred_bbox','pred_format','confidence','gt_size','iou']
"""
# imports
import argparse
import os
import pandas as pd
import numpy as np
parser = argparse.ArgumentParser(description='Parser')
parser.add_argument('--GT_bbox_Dir', type=str, help='directory with txt files containing GT bboxes')
parser.add_argument('--Pred_bbox_Dir', type=str, help='directory with txt files containing predicted bboxes')
parser.add_argument('--outptuFile', type=str, help='filepath to results_{set}_{model}.csv to write')
parser.add_argument('--gt_format', type=str, default='xywh_pix', help='txt format for GT bbox files')
parser.add_argument('--pred_format', type=str, default='y1x1y2x2_pix', help='txt format for predicted bbox files')
args = parser.parse_args()
# import required modules
import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
#from shapely.geometry import box
import random
from sklearn import metrics
import humanfriendly
import time
# Utils
def parse_txt (label_fp, format_bbox, gt_or_pred):
"""Obtain x_min, y_min, x_max, and y_max of bounding box from txt file
Args:
label_fp (str): filepath to bounding box .txt file in detect.py output format
format_bbox: 'x1y1x2y2' or 'xywh_pix' or 'x1y1x2y2_pix'
gt_or_pred: 'predicted' or 'ground_truth'
Returns:
coords (numpy array)
conf (float, returned only if dataset == 'predicted')
"""
if format_bbox == 'y1x1y2x2_pix':
with open(label_fp, 'r') as label:
line = str(label.readline())
if len(line) == 0:
return None
else:
vals = line.split(' ')
x_min = int(float(vals[1]) * 419)
y_min = int(float(vals[2]) * 419)
x_max = int(float(vals[3]) * 419)
y_max = int(float(vals[0]) * 419)
coords = np.array([y_min, x_min, y_max, x_max])
if gt_or_pred == 'ground_truth':
return coords
elif gt_or_pred == 'predicted':
conf = float(vals[5])
conf = '%.4f'%(conf)
return coords, conf
elif format_bbox == 'xywh_pix':
with open(label_fp, 'r') as label:
line = str(label.readline())
if len(line) == 0:
return None
else:
vals = line.split(' ')
norm_x = float(vals[1])
norm_y = float(vals[2])
norm_w = float(vals[3])
norm_h = float(vals[4])
x_min = int((norm_x * 419) - ((norm_w * 419) / 2))
y_min = int((norm_y * 419) - ((norm_h * 419) / 2))
x_max = int((norm_x * 419) + ((norm_w * 419) / 2))
y_max = int((norm_y * 419) + ((norm_h * 419) / 2))
coords = np.array([x_min, y_min, x_max, y_max])
if gt_or_pred == 'ground_truth':
return coords
elif gt_or_pred == 'predicted':
conf = float(vals[5])
conf = '%.4f'%(conf)
return coords, conf
def calc_IoU (bb1, bb2, gt_format, pred_format):
"""
Calculate the Intersection over Union (IoU) of two bounding boxes.
Adapted from: https://stackoverflow.com/questions/25349178/calculating-percentage-of-bounding-box-overlap-for-image-detector-evaluation
Args:
bb1: [x1,y1,x2,y2]
bb2: [x1,y1,x2,y2]
The (x1, y1) position is at the top left corner (or the bottom right - either way works).
The (x2, y2) position is at the bottom right corner (or the top left).
Returns:
intersection_over_union, a float in [0, 1]
"""
# convert to x1y1x2y2 format if needed
if gt_format == 'y1x1y2x2_pix':
y_max, x_min, y_min, x_max = bb1
bb1 = [x_min, y_min, x_max, y_max]
if pred_format == 'y1x1y2x2_pix':
y_max, x_min, y_min, x_max = bb2
bb2 = [x_min, y_min, x_max, y_max]
# Determine the coordinates of the intersection rectangle
x_left = max(bb1[0], bb2[0])
y_top = max(bb1[1], bb2[1])
x_right = min(bb1[2], bb2[2])
y_bottom = min(bb1[3], bb2[3])
if x_right < x_left or y_bottom < y_top:
return 0.0
# The intersection of two axis-aligned bounding boxes is always an
# axis-aligned bounding box
intersection_area = (x_right - x_left) * (y_bottom - y_top)
# Compute the area of both AABBs
bb1_area = (bb1[2] - bb1[0]) * (bb1[3] - bb1[1])
bb2_area = (bb2[2] - bb2[0]) * (bb2[3] - bb2[1])
# Compute the intersection over union by taking the intersection
# area and dividing it by the sum of prediction + ground-truth
# areas - the intersection area.
iou = intersection_area / float(bb1_area + bb2_area - intersection_area)
assert iou >= 0.0
assert iou <= 1.0
return iou
def make_results_table (gt_bbox_dir, pred_bbox_dir, outputFile,
gt_format='xywh_pix', pred_format='y1x1y2x2_pix'):
'''
Creates dataframe from directories and format for ground truth boxes and predicted boxes.
Args:
gt_bbox_dir: directory of ground truth bounding boxes
pred_bbox_dir: directory of ground truth bounding boxes
outputFile : filepath to .csv results table to write (results_{set}_{model}.csv)
gt_format (Optional): format of ground truth bounding boxes 'y1x1y2x2_pix' or 'xywh_pix'. Default is 'xywh_pix'
pred_format (Optional): format of ground truth bounding boxes 'y1x1y2x2_pix' or 'xywh_pix' . Default is 'y1x1y2x2_pix'
Returns:
dataframe with cols:
['img_id','gt_bbox','gt_format','pred_bbox','pred_format','confidence','gt_size','iou']
'''
print('Loading GT & predicted bboxes...')
startTime = time.time()
# collect list of filenames
pred_bbox_fn = [fn[:-4] for fn in os.listdir(pred_bbox_dir)]
gt_bbox_fn = [fn[:-4] for fn in os.listdir(gt_bbox_dir)]
elapsed = time.time() - startTime
print("Finished loaded {} GT bboxes and {} predicted bboxes in {}".format(len(gt_bbox_fn),len(pred_bbox_fn),
humanfriendly.format_timespan(elapsed)))
# create dictionaries of ground_truth bboxes, predicted bboxes, and confidence
ground_truth_dict = {}
for fn in gt_bbox_fn:
bbox = parse_txt(os.path.join(gt_bbox_dir, fn+'.txt'), gt_format, 'ground_truth')
ground_truth_dict[fn] = bbox
predicted_dict = {}
conf_dict = {}
for fn in pred_bbox_fn:
bbox, conf = parse_txt(os.path.join(pred_bbox_dir, fn+'.txt'), pred_format, 'predicted')
predicted_dict[fn] = bbox
conf_dict[fn] = conf
# build dataframe of image, ground_truth bboxes, predicted bboxes, and confidence
print('Creating results table with ...')
data = {'img_id' : gt_bbox_fn[:-4]}
detect_df = pd.DataFrame(data, columns=['img_id'])
print('... GT bboxes')
detect_df['gt_bbox'] = detect_df['img_id'].map(pd.Series(ground_truth_dict))
detect_df['gt_format'] = gt_format
print('... predicted bboxes')
detect_df['pred_bbox'] = detect_df['img_id'].map(pd.Series(predicted_dict))
detect_df['pred_format'] = pred_format
print('... confidence')
detect_df['confidence'] = detect_df['img_id'].map(pd.Series(conf_dict))
# add ground_truth bbox size to dataframe
print('... Bbox size')
detect_df['gt_size'] = detect_df.apply(
lambda row: None if row.gt_bbox is None else int((row.gt_bbox[2] - row.gt_bbox[0]) * (row.gt_bbox[3] - row.gt_bbox[1])),
axis=1
)
# add IoU - cameratraps to dataframe
print('... IoU')
iou_dict = {}
for fn in predicted_dict:
if 'not_a_dam' not in fn:
iou = calc_IoU(ground_truth_dict[fn], predicted_dict[fn], gt_format, pred_format)
iou_dict[fn] = iou
else:
iou_dict[fn] = None
detect_df['iou'] = detect_df['img_id'].map(pd.Series(iou_dict))
print('... Done \n')
# set index to img_id
detect_df = detect_df.set_index('img_id')
print('Writing CSV ...')
detect_df.to_csv(outputFile)
print('... Done')
return detect_df
#
if __name__ == '__main__':
make_results_table(args.GT_bbox_Dir,
args.Pred_bbox_Dir,
args.outputFile,
args.gt_format,
args.pred_format)