-
Notifications
You must be signed in to change notification settings - Fork 3
/
utils.py
235 lines (200 loc) · 6.89 KB
/
utils.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
import cupy as cp
import faiss
import numpy as np
import torch
import torch.nn.functional as F
from cupyx.scipy.sparse import csr_matrix, diags, eye
from cupyx.scipy.sparse import linalg as s_linalg
def search_faiss(X, Q, k):
res = faiss.StandardGpuResources()
res.setDefaultNullStreamAllDevices()
res.setTempMemory(0)
s, knn = faiss.knn_gpu(res, Q, X, k, metric=faiss.METRIC_INNER_PRODUCT)
return knn, s
def normalize_connection_graph(G):
W = csr_matrix(G)
W = W - diags(W.diagonal(), 0)
S = W.sum(axis=1)
S[S == 0] = 1
D = cp.array(1.0 / cp.sqrt(S))
D[cp.isnan(D)] = 0
D[cp.isinf(D)] = 0
D_mh = diags(D.reshape(-1), 0)
Wn = D_mh * W * D_mh
return Wn
def knn2laplacian(knn, s, alpha=0.99):
N = knn.shape[0]
k = knn.shape[1]
row_idx = np.arange(N)
row_idx_rep = np.tile(row_idx, (k, 1)).T
knn_flat = knn.flatten("F")
row_idx_rep_flat = row_idx_rep.flatten("F")
sim_flat = s.flatten("F")
valid_knn = np.where(knn_flat != -1)[0]
knn_flat = cp.array(knn_flat[valid_knn])
row_idx_rep_flat = cp.array(row_idx_rep_flat[valid_knn])
sim_flat = cp.array(sim_flat[valid_knn])
W = csr_matrix(
(sim_flat, (row_idx_rep_flat, knn_flat)),
shape=(N, N),
)
W = W + W.T
Wn = normalize_connection_graph(W)
L = eye(Wn.shape[0]) - alpha * Wn
return L
def dfs_search(L, Y, tol=1e-6, maxiter=50, cast_to_numpy=True):
out = s_linalg.cg(L, Y, tol=tol, maxiter=maxiter)[0]
if cast_to_numpy:
return cp.asnumpy(out)
else:
return out
def normalize(x):
return F.normalize(torch.tensor(x), p=2, dim=1).cpu().numpy()
def accuracy(scores, labels):
preds = np.argmax(scores, axis=1)
acc = np.mean(100.0 * (preds == labels))
return acc
def get_data(dataset, model="RN50"):
try:
train_features = np.load(f"features/{dataset}/{model}_train_feats.npy")
train_features = normalize(train_features.astype(np.float32))
train_targets = np.load(f"features/{dataset}/{model}_train_targets.npy")
except OSError:
print("No train features! Inductive setting will not be possible!")
train_features = None
train_targets = None
try:
val_features = np.load(f"features/{dataset}/{model}_val_feats.npy")
val_features = normalize(val_features.astype(np.float32))
val_targets = np.load(f"features/{dataset}/{model}_val_targets.npy")
except OSError:
print("No val features!!!")
val_features = None
val_targets = None
try:
test_features = np.load(f"features/{dataset}/{model}_test_feats.npy")
test_features = normalize(test_features.astype(np.float32))
test_targets = np.load(f"features/{dataset}/{model}_test_targets.npy")
except OSError:
print("No test features! Using val features as test!")
if val_features is None:
raise ValueError("No val features either!")
test_features = val_features
test_targets = val_targets
val_features = None
val_targets = None
try:
clf_text = np.load(
f"features/{dataset}/classifiers/{model}_text_classifier.npy"
)
clf_text = normalize(clf_text.T)
except OSError:
raise ValueError("No extracted text classifier!")
try:
clf_cupl_text = np.load(
f"features/{dataset}/classifiers/{model}_cupl_text_classifier.npy"
)
clf_cupl_text = normalize(clf_cupl_text.T)
except OSError:
clf_cupl_text = None
try:
clf_image_train = np.load(
f"features/{dataset}/classifiers/{model}_inmap_proxy_classifier_train.npy"
)
clf_image_train = normalize(clf_image_train.T)
except OSError:
clf_image_train = None
try:
clf_cupl_image_train = np.load(
f"features/{dataset}/classifiers/{model}_cupl_inmap_proxy_classifier_train.npy"
)
clf_cupl_image_train = normalize(clf_cupl_image_train.T)
except OSError:
clf_cupl_image_train = None
try:
clf_image_val = np.load(
f"features/{dataset}/classifiers/{model}_inmap_proxy_classifier_val.npy"
)
clf_image_val = normalize(clf_image_val.T)
except OSError:
clf_image_val = None
try:
clf_cupl_image_val = np.load(
f"features/{dataset}/classifiers/{model}_cupl_inmap_proxy_classifier_val.npy"
)
clf_cupl_image_val = normalize(clf_cupl_image_val.T)
except OSError:
clf_cupl_image_val = None
try:
clf_image_test = np.load(
f"features/{dataset}/classifiers/{model}_inmap_proxy_classifier_test.npy"
)
clf_image_test = normalize(clf_image_test.T)
except OSError:
print(
"No InMaP classifer learned on test, so using the one trained on val instead!"
)
clf_image_test = clf_image_val
try:
clf_cupl_image_test = np.load(
f"features/{dataset}/classifiers/{model}_cupl_inmap_proxy_classifier_test.npy"
)
clf_cupl_image_test = normalize(clf_cupl_image_test.T)
except OSError:
print(
"No InMaP classifer learned on test, so using the one trained on val instead!"
)
clf_cupl_image_test = clf_cupl_image_val
return (
train_features,
train_targets,
val_features,
val_targets,
test_features,
test_targets,
clf_text,
clf_image_train,
clf_image_val,
clf_image_test,
clf_cupl_text,
clf_cupl_image_train,
clf_cupl_image_val,
clf_cupl_image_test,
)
def voc_ap(rec, prec, true_num):
mrec = np.concatenate(([0.0], rec, [1.0]))
mpre = np.concatenate(([0.0], prec, [0.0]))
for i in range(mpre.size - 1, 0, -1):
mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i])
i = np.where(mrec[1:] != mrec[:-1])[0]
ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])
return ap
def voc_mAP(imagessetfilelist, num, return_each=False):
seg = imagessetfilelist
gt_label = seg[:, num:].astype(np.int32)
sample_num = len(gt_label)
class_num = num
tp = np.zeros(sample_num)
fp = np.zeros(sample_num)
aps = []
for class_id in range(class_num):
confidence = seg[:, class_id]
sorted_ind = np.argsort(-confidence)
sorted_label = [gt_label[x][class_id] for x in sorted_ind]
for i in range(sample_num):
tp[i] = sorted_label[i] > 0
fp[i] = sorted_label[i] <= 0
true_num = 0
true_num = sum(tp)
fp = np.cumsum(fp)
tp = np.cumsum(tp)
rec = tp / float(true_num)
prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
ap = voc_ap(rec, prec, true_num)
aps += [ap]
np.set_printoptions(precision=6, suppress=True)
aps = np.array(aps) * 100
mAP = np.mean(aps)
if return_each:
return mAP, aps
return mAP