-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
212 lines (167 loc) · 7.68 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
from torchvision import transforms, ops
import torch.nn.functional as F
import torch
import numpy as np
class FilterTools():
def __init__(self, short_mems, long_mems):
self.short_mems = short_mems
self.long_mems = long_mems
self.best_conf = None
self.best_emb = None
self.target_mem = None
def feature_sim_from_gdino(self, dets, feature, ref_pos, ref_conf):
rh, rw = feature.tensors.shape[2] / 1080 , feature.tensors.shape[3] / 1920
det_feats = []
cropped_feats = []
for i, det in enumerate(dets):
xc = int((det[0] + det[2])/2 * rw)
yc = int((det[1] + det[3])/2 * rh)
det_feats.append(feature.tensors[:, :, yc, xc])
if i == ref_pos:
x, y, xx, yy = det
if xx - x == 0:
xx += 1
if yy - y == 0:
yy += 1
roi = feature.tensors[:, :, int(y * rh):int(yy * rh),
int((x + (xx-x)/8) * rw):int(xx * rw)]
roi = torch.nn.functional.interpolate(roi, size=(1, 1), mode='bilinear', align_corners=True)
roi.squeeze()
cropped_feats.append(roi)
roi = feature.tensors[:, :, int(y * rh):int(yy * rh),
int(x * rw):int((xx - (xx-x)/8) * rw)]
roi = torch.nn.functional.interpolate(roi, size=(1, 1), mode='bilinear', align_corners=True)
roi.squeeze()
cropped_feats.append(roi)
roi = feature.tensors[:, :, int((y + (yy-y)/8)* rh):int(yy * rh),
int(x * rw):int(xx * rw)]
roi = torch.nn.functional.interpolate(roi, size=(1, 1), mode='bilinear', align_corners=True)
roi.squeeze()
cropped_feats.append(roi)
roi = feature.tensors[:, :, int(y * rh):int((yy - (yy-y)/8)* rh),
int(x * rw):int(xx * rw)]
roi = torch.nn.functional.interpolate(roi, size=(1, 1), mode='bilinear', align_corners=True)
roi.squeeze()
cropped_feats.append(roi)
# # RoI Align
# x, y, xx, yy = det
# if xx - x == 0:
# xx += 1
# if yy - y == 0:
# yy += 1
# roi = feature.tensors[:, :, int(y * rh):int(yy * rh),
# int(x * rw):int(xx * rw)]
# roi = torch.nn.functional.interpolate(roi, size=(1, 1), mode='bicubic', align_corners=True)
# roi.squeeze()
# det_feats.append(roi)
cropped_embs = torch.squeeze(torch.stack(cropped_feats, dim=0))
embs = torch.squeeze(torch.stack(det_feats, dim=0), 1)
ref_emb = embs[ref_pos].unsqueeze(0)
if self.target_mem == None:
self.target_mem = ref_emb
if self.long_mems > 0:
self.best_conf = np.array([ref_conf])
self.best_emb = ref_emb
self.cropped_embs = cropped_embs
else:
if self.long_mems > 0:
if (self.best_emb.size()[0] < self.long_mems) and (self.best_conf.min() <= ref_conf):
np.append(self.best_conf , ref_conf)
self.best_emb = torch.cat((self.best_emb, ref_emb), dim=0)
else:
if self.best_conf.min() <= ref_conf:
min_idx = self.best_conf.argmin()
self.best_conf[min_idx] = ref_conf
self.best_emb[min_idx] = ref_emb
if self.best_conf.max() <= ref_conf:
self.cropped_embs = cropped_embs
if self.target_mem.size()[0] < self.short_mems:
self.target_mem = torch.cat((self.target_mem, ref_emb), dim=0)
elif self.target_mem.size()[0] == self.short_mems:
self.target_mem = torch.cat((self.target_mem[1:, :], ref_emb), dim=0)
t1_norm = F.normalize(embs, dim=1)
t2_norm = F.normalize(self.target_mem, dim=1)
result = torch.mean(torch.mm(t1_norm, t2_norm.t()), dim=1)
if self.long_mems:
t3_norm = F.normalize(self.best_emb, dim=1)
best_res = torch.mean(torch.mm(t1_norm, t3_norm.t()), dim=1)
t4_norm = F.normalize(self.cropped_embs, dim=1)
cropped_res = torch.mean(torch.mm(t1_norm, t4_norm.t()), dim=1)
else:
best_res = None
cropped_res = None
return result, best_res, cropped_res, embs
def nms(bounding_boxes, confidence_score, threshold=0.6):
# If no bounding boxes, return empty list
if len(bounding_boxes) == 0:
return [], []
# Bounding boxes
boxes = bounding_boxes
# coordinates of bounding boxes
start_x = boxes[:, 0]
start_y = boxes[:, 1]
end_x = boxes[:, 2]
end_y = boxes[:, 3]
# Confidence scores of bounding boxes
score = np.array(confidence_score)
# Picked bounding boxes
picked_boxes = []
picked_score = []
# Compute areas of bounding boxes
areas = (end_x - start_x + 1) * (end_y - start_y + 1)
# Sort by confidence score of bounding boxes
order = np.argsort(score)
# Iterate bounding boxes
while order.size > 0:
# The index of largest confidence score
index = order[-1]
# Pick the bounding box with largest confidence score
picked_boxes.append(bounding_boxes[index])
picked_score.append(confidence_score[index])
# Compute ordinates of intersection-over-union(IOU)
x1 = np.maximum(start_x[index], start_x[order[:-1]])
x2 = np.minimum(end_x[index], end_x[order[:-1]])
y1 = np.maximum(start_y[index], start_y[order[:-1]])
y2 = np.minimum(end_y[index], end_y[order[:-1]])
# Compute areas of intersection-over-union
w = np.maximum(0.0, x2 - x1 + 1)
h = np.maximum(0.0, y2 - y1 + 1)
intersection = w * h
# Compute the ratio between intersection and union
ratio = intersection / (areas[index] + areas[order[:-1]] - intersection)
left = np.where(ratio < threshold)
order = order[left]
return np.array(picked_boxes), np.array(picked_score)
def cal_iou(bbox1, bbox2, mode='iou'):
# Tính toán tọa độ của vùng giao nhau (intersection)
x1 = max(bbox1[0], bbox2[0])
y1 = max(bbox1[1], bbox2[1])
x2 = min(bbox1[2], bbox2[2])
y2 = min(bbox1[3], bbox2[3])
# Tính toán diện tích của vùng giao nhau
intersection_area = max(0, x2 - x1 + 1) * max(0, y2 - y1 + 1)
# Tính toán diện tích tổng của hai bbox
bbox1_area = (bbox1[2] - bbox1[0] + 1) * (bbox1[3] - bbox1[1] + 1)
bbox2_area = (bbox2[2] - bbox2[0] + 1) * (bbox2[3] - bbox2[1] + 1)
# Tính toán IoU
if mode == 'iou':
iou = intersection_area / float(bbox1_area + bbox2_area - intersection_area)
else:
#bbox1 nên là box lớn hơn, nếu ko lớn hơn thì kqua sẽ không đủ cao
iou = intersection_area / bbox2_area
return iou
def contains_bbox(bbox1, bbox2):
"""
Returns True if bbox2 is contained inside bbox1, False otherwise.
Args:
bbox1: The larger bounding box.
bbox2: The smaller bounding box.
Returns:
True if bbox2 is contained inside bbox1, False otherwise.
"""
top_left_1 = (bbox1[0], bbox1[1])
bottom_right_1 = (bbox1[2], bbox1[3])
top_left_2 = (bbox2[0], bbox2[1])
bottom_right_2 = (bbox2[2], bbox2[3])
return top_left_2[0] >= top_left_1[0] and top_left_2[1] >= top_left_1[1] \
and bottom_right_2[0] <= bottom_right_1[0] and bottom_right_2[1] <= bottom_right_1[1]