-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathevaluate.py
279 lines (216 loc) · 8.93 KB
/
evaluate.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
273
274
275
276
277
278
279
# Copyright (C) 2022 * Ltd. All rights reserved.
# author : Sanghyun Jo <[email protected]>
import os
import sys
import cv2
import ray
import tqdm
import torch
import numpy as np
from PIL import Image
from core import datasets
from tools.ai import evaluators, augment_utils
from tools.general import io_utils, json_utils, cv_utils, pickle_utils
@ray.remote
def update_mIoU(obj, pred_cam, gt_mask, label):
# 1. define numpy
meter_dict = {}
for th in obj.thresholds:
meter_dict[th] = {
'P' : np.zeros(obj.num_classes, dtype=np.float32),
'T' : np.zeros(obj.num_classes, dtype=np.float32),
'TP' : np.zeros(obj.num_classes, dtype=np.float32),
'FP_BG' : np.zeros(obj.num_classes, dtype=np.float32),
'FP_FG' : np.zeros(obj.num_classes, dtype=np.float32),
}
# 2. calculate P, T, and TP
for th in obj.thresholds:
pred_mask = np.pad(pred_cam, ((1, 0), (0, 0), (0, 0)), mode='constant', constant_values=th)
pred_mask = np.argmax(pred_mask, axis=0)
obj_mask = gt_mask != obj.ignore_index
correct_mask = (pred_mask==gt_mask) * obj_mask
if obj.detail:
fp_mask = (pred_mask != gt_mask) * obj_mask
bg_mask = (gt_mask == 0) * obj_mask
fg_mask = (gt_mask > 0) * obj_mask
for i in range(obj.num_classes):
if i > 0 and label is not None:
if label[i - 1] == 0:
continue
meter_dict[th]['P'][i] += np.sum((pred_mask==i)*obj_mask)
meter_dict[th]['T'][i] += np.sum((gt_mask==i)*obj_mask)
meter_dict[th]['TP'][i] += np.sum((gt_mask==i)*correct_mask)
if i > 0 and obj.detail:
meter_dict[th]['FP_BG'][i] += np.sum((pred_mask==i)*fp_mask*bg_mask)
meter_dict[th]['FP_FG'][i] += np.sum((pred_mask==i)*fp_mask*fg_mask)
meter_dict['label'] = label
return meter_dict
@ray.remote
def update_mIoU_for_semantic_segmentation(obj, pred_mask, gt_mask):
# 1. define numpy
meter_dict = {
'P' : np.zeros(obj.num_classes, dtype=np.float32),
'T' : np.zeros(obj.num_classes, dtype=np.float32),
'TP' : np.zeros(obj.num_classes, dtype=np.float32),
'FP_BG' : np.zeros(obj.num_classes, dtype=np.float32),
'FP_FG' : np.zeros(obj.num_classes, dtype=np.float32),
}
if len(pred_mask.shape) == 3:
pred_mask = np.argmax(pred_mask, axis=0)
obj_mask = gt_mask != obj.ignore_index
correct_mask = (pred_mask == gt_mask) * obj_mask
if obj.detail:
fp_mask = (pred_mask != gt_mask) * obj_mask
bg_mask = (gt_mask == 0) * obj_mask
fg_mask = (gt_mask > 0) * obj_mask
for i in range(obj.num_classes):
meter_dict['P'][i] += np.sum((pred_mask==i)*obj_mask)
meter_dict['T'][i] += np.sum((gt_mask==i)*obj_mask)
meter_dict['TP'][i] += np.sum((gt_mask==i)*correct_mask)
if i > 0 and obj.detail:
meter_dict['FP_BG'][i] += np.sum((pred_mask==i)*fp_mask*bg_mask)
meter_dict['FP_FG'][i] += np.sum((pred_mask==i)*fp_mask*fg_mask)
return meter_dict
def main(args):
# set directories
pred_dir = args.pred_dir + f'{args.folder}/{args.tag}/{args.domain}/'
if args.detail:
analysis_dir = io_utils.create_directory('./experiments/analysis/per-class/')
analysis_path = analysis_dir + args.tag + '.json'
log_func = lambda string='': print(string)
# read dataset information
data_dict = json_utils.read_json(f'./data/{args.dataset}.json')
# for datasets
test_transform = augment_utils.Compose(
[
augment_utils.Normalize(),
augment_utils.Transpose(),
]
)
test_dataset = datasets.Dataset_For_Analysis(
args.root_dir,
args.domain,
test_transform,
name=args.dataset,
single=args.single
)
# for evaluation
colors = cv_utils.get_colors(data_dict)
denorm_fn = augment_utils.Denormalize()
if args.parallel or 'predictions' in args.folder:
ids = []
ray.init(num_cpus=args.num_workers)
if 'predictions' in args.folder:
evaluator_for_cam = evaluators.Evaluator_For_CAM(data_dict['class_names'], ignore_index=255, st_th=args.st_th, end_th=args.end_th, th_interval=args.th_interval)
else:
evaluator_for_ss = evaluators.Evaluator_For_Semantic_Segmentation(data_dict['class_names'], ignore_index=255)
print(evaluator_for_ss.class_names)
if args.debug:
count = 0
for image_id, image, label, gt_mask in tqdm.tqdm(test_dataset):
gt_mask = np.asarray(gt_mask)
if args.debug:
count += 1
if count >= 2500:
break
# if not os.path.isfile(pred_dir + image_id + '.pkl'):
# continue
if 'predictions' in args.folder:
infer_dict = pickle_utils.load_pickle(pred_dir + image_id + '.pkl')
# print()
# print(infer_dict['keys'][1:]-1)
# print(infer_dict['seed'].shape)
# print()
_, h, w = image.shape
pred = np.zeros((len(label), h, w), dtype=np.float32)
if len(infer_dict['keys'][1:]) > 0:
pred[infer_dict['keys'][1:]-1] = infer_dict['seed']
"""
# evaluator_for_cam.add([pred, gt_mask, label])
evaluator_for_cam.add([pred, gt_mask, None])
"""
id = update_mIoU.remote(
evaluator_for_cam,
pred,
gt_mask,
label if args.domain == 'train' else None
)
ids.append(id)
if len(ids) > (args.num_workers * 10):
for meter_dict in ray.get(ids):
evaluator_for_cam.add_from_data(
label=meter_dict['label'],
meter_dict=meter_dict
)
ids = []
else:
pred_mask = Image.open(pred_dir + image_id + '.png')
pred_mask = np.asarray(pred_mask)
# print(image_id)
if args.parallel:
id = update_mIoU_for_semantic_segmentation.remote(
evaluator_for_ss,
pred_mask,
gt_mask,
)
ids.append(id)
# print(len(ids), args.num_workers * 10)
if len(ids) > (args.num_workers * 10):
for meter_dict in ray.get(ids):
evaluator_for_ss.add_from_data(meter_dict)
# print(evaluator_for_ss.meter_dict['P'][0])
# print(evaluator_for_ss.meter_dict['T'][0])
# print(evaluator_for_ss.meter_dict['TP'][0])
# print()
ids = []
else:
evaluator_for_ss.add([pred_mask, gt_mask])
# if args.debug:
# cv2.imshow('Ground Truth', colors[gt_mask])
# cv2.imshow('Prediction', cv_utils.apply_colormap(pred))
# # cv2.imshow('Prediction', colors[pred_mask])
# cv2.waitKey(0)
if 'predictions' in args.folder:
if len(ids) > 0:
for meter_dict in ray.get(ids):
evaluator_for_cam.add_from_data(
label=meter_dict['label'],
meter_dict=meter_dict
)
evaluator_for_cam.print(args.tag)
else:
if args.parallel and len(ids) > 0:
for meter_dict in ray.get(ids):
evaluator_for_ss.add_from_data(meter_dict)
if args.detail:
IoU_list = evaluator_for_ss.print_with_detail()
class_names = evaluator_for_ss.class_names
print(len(class_names), len(IoU_list), np.mean(IoU_list))
IoU_dict = {
name: float(IoU) for IoU, name in zip(IoU_list, class_names)
}
IoU_dict['mIoU'] = np.mean(IoU_list)
json_utils.write_json(analysis_path, IoU_dict)
else:
evaluator_for_ss.print(args.tag)
if __name__ == '__main__':
parser = io_utils.Parser()
# environment
parser.add('gpus', '0', str)
parser.add('num_workers', 16, int)
# dataset
parser.add('dataset', 'VOC', str)
parser.add('root_dir', '../VOC2012/', str)
parser.add('pred_dir', './experiments/', str)
parser.add('domain', 'train', str)
parser.add('single', False, bool)
# evaluation configuration
parser.add('tag', 'ResNet50@RS(ep=5, cam=1.0, seg=1.0)+CF(c=0.55, u=0.10)+CM(ep=15, cls=1.0, cam=1.0, seg=1.0)', str)
parser.add('folder', 'pseudo-labels', str)
parser.add('debug', False, bool)
parser.add('parallel', False, bool)
parser.add('detail', False, bool)
parser.add('st_th', 0.10, float)
parser.add('end_th', 0.80, float)
parser.add('th_interval', 0.05, float)
main(parser.get_args())