-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluate_SOR.py
295 lines (204 loc) · 7.74 KB
/
evaluate_SOR.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
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
from DatasetTest import DatasetTest
from sklearn.metrics import mean_absolute_error
import cv2
import numpy as np
import pickle
import os
import utils
import scipy.stats as sc
WIDTH = 640
HEIGHT = 480
# Percentage of object pixels having predicted saliency value to consider as salient object
# for cases where object segments overlap each other
SEG_THRESHOLD = .5
def load_saliency_map(path):
# Load mask
sal_map = cv2.imread(path, 1).astype(np.float32)
# Need only one channel
sal_map = sal_map[:, :, 0]
# Normalize to 0-1
sal_map /= 255.0
return sal_map
def eval_mae(dataset, map_path):
print("Calculating MAE...")
mae_list = []
num = len(dataset.img_ids)
for i in range(num):
image_id = dataset.img_ids[i]
p = map_path + image_id + ".png"
pred_mask = load_saliency_map(p)
gt_mask = dataset.load_gt_mask(image_id)
# Flatten masks
gt_mask = gt_mask.flatten()
pred_mask = pred_mask.flatten()
mae = mean_absolute_error(gt_mask, pred_mask)
mae_list.append(mae)
print("\n")
avg_mae = sum(mae_list) / len(mae_list)
print("Average MAE Images = ", avg_mae)
def eval_mae_binary_mask(dataset, map_path):
print("Calculating MAE (Binary Saliency)...")
mae_list = []
num = len(dataset.img_ids)
for i in range(num):
image_id = dataset.img_ids[i]
p = map_path + image_id + ".png"
pred_mask = load_saliency_map(p)
gt_mask = dataset.load_gt_mask(image_id)
# Convert masks to binary
pred_mask[pred_mask > 0] = 1
gt_mask[gt_mask > 0] = 1
# Flatten masks
gt_mask = gt_mask.flatten()
pred_mask = pred_mask.flatten()
mae = mean_absolute_error(gt_mask, pred_mask)
mae_list.append(mae)
print("\n")
avg_mae = sum(mae_list) / len(mae_list)
print("Average MAE Images (Binary Masks) = ", avg_mae)
def calculate_spr(dataset, model_pred_data_path, out_path):
print("Calculating SOR...")
# Load GT Rank
gt_rank_order = dataset.gt_rank_orders
spr_data = []
num = len(dataset.img_ids)
for i in range(num):
# Image Id
image_id = dataset.img_ids[i]
print("\n")
print(i + 1, " / ", num, " - ", image_id)
# ********** Dataset information
idx = dataset.img_ids.index(image_id)
sal_obj_idx = dataset.sal_obj_idx_list[idx]
N = len(sal_obj_idx)
# load seg data
obj_seg = dataset.obj_seg[i]
instance_masks = []
instance_pix_count = []
# print(sal_obj_idx)
# print(obj_seg)
# print(len(sal_obj_idx))
# print(len(obj_seg))
# Create mask for each salient object
for s_i in range(len(sal_obj_idx)):
sal_idx = sal_obj_idx[s_i]
# Get corresponding segmentation data
seg = obj_seg[sal_idx]
# Binary mask of object segment
mask = utils.get_obj_mask(seg, HEIGHT, WIDTH)
# Count number of pixels of object segment
pix_count = mask.sum()
instance_masks.append(mask)
instance_pix_count.append(pix_count)
# ********** Load Predicted Rank
pred_data_path = model_pred_data_path + dataset.img_ids[i] + ".png"
pred_sal_map = cv2.imread(pred_data_path)[:, :, 0]
# Get corresponding predicted rank for each gt salient objects
pred_ranks = []
# Create mask for each salient object
for s_i in range(len(instance_masks)):
gt_seg_mask = instance_masks[s_i]
gt_pix_count = instance_pix_count[s_i]
pred_seg = np.where(gt_seg_mask == 1, pred_sal_map, 0)
# number of pixels with predicted values
pred_pix_loc = np.where(pred_seg > 0)
pred_pix_num = len(pred_pix_loc[0])
# Get rank of object
r = 0
if pred_pix_num > int(gt_pix_count * SEG_THRESHOLD):
vals = pred_seg[pred_pix_loc[0], pred_pix_loc[1]]
mode = sc.mode(vals)[0][0]
r = mode
pred_ranks.append(r)
# ********** Load GT Rank
gt_rank_order_list = gt_rank_order[i]
# Get Gt Rank Order of salient objects
gt_ranks = []
for j in range(N):
s_idx = sal_obj_idx[j]
gt_r = gt_rank_order_list[s_idx]
gt_ranks.append(gt_r)
# Remove objects with no saliency value in both list
gt_ranks, pred_ranks, use_indices_list = \
utils.get_usable_salient_objects_agreed(gt_ranks, pred_ranks)
spr = None
if len(gt_ranks) > 1:
spr = sc.spearmanr(gt_ranks, pred_ranks)
# spr = sc.pearsonr(gt_ranks, pred_ranks)
elif len(gt_ranks) == 1:
spr = 1
d = [image_id, spr, use_indices_list]
spr_data.append(d)
with open(out_path, "wb") as f:
pickle.dump(spr_data, f)
def extract_spr_value(data_list):
use_idx_list = []
spr = []
for i in range(len(data_list)):
s = data_list[i][1]
if s == 1:
spr.append(s)
use_idx_list.append(i)
elif s and not np.isnan(s[0]):
spr.append(s[0])
use_idx_list.append(i)
return spr, use_idx_list
def cal_avg_spr(data_list):
spr = np.array(data_list)
avg = np.average(spr)
return avg
def get_norm_spr(spr_value):
# m - r_min
# m -> ---------------- x (t_max - t_min) + t_min
# r_max - r_min
#
# m = measure value
# r_min = min range of measurement
# r_max = max range of measurement
# t_min = min range of desired scale
# t_max = max range of desired scale
r_min = -1
r_max = 1
norm_spr = (spr_value - r_min) / (r_max - r_min)
return norm_spr
def eval_spr(spr_data_path):
with open(spr_data_path, "rb") as f:
spr_all_data = pickle.load(f)
spr_data, spr_use_idx = extract_spr_value(spr_all_data)
pos_l = []
neg_l = []
for i in range(len(spr_data)):
if spr_data[i] > 0:
pos_l.append(spr_data[i])
else:
neg_l.append(spr_data[i])
print("Positive SPR: ", pos_l)
print("Negative SPR: ", neg_l)
print("Positive SPR: ", len(pos_l))
print("Negative SPR: ", len(neg_l))
avg_spr = cal_avg_spr(spr_data)
avg_spr_norm = get_norm_spr(avg_spr)
print("\n----------------------------------------------------------")
print("Data path: ", spr_data_path)
print(len(spr_data), "/", len(spr_all_data), " - ", (len(spr_all_data) - len(spr_data)), "Images Not used")
print("Average SPR Saliency: ", avg_spr)
print("Average SPR Saliency Normalized: ", avg_spr_norm)
if __name__ == '__main__':
print("Evaluate")
DATASET_ROOT = "E:/Datasets/ASSR/" # Change to your location
PRE_PROC_DATA_ROOT = "E:/Datasets/ASSR_Data/" # Change to your location
data_split = "test"
dataset = DatasetTest(DATASET_ROOT, PRE_PROC_DATA_ROOT, data_split, eval_spr=True)
####################################################
map_path = "E:/Datasets/ASSR/results/final-gray/"
# Calculate MAE
eval_mae(dataset, map_path)
eval_mae_binary_mask(dataset, map_path)
####################################################
out_root = "../spr_data/"
out_path = out_root + "spr_data"
if not os.path.exists(out_root):
os.makedirs(out_root)
# Calculate SOR
calculate_spr(dataset, map_path, out_path)
eval_spr(out_path)