-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathyolov2_loss.py
141 lines (106 loc) · 6.38 KB
/
yolov2_loss.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
import tensorflow as tf
def bbox_ious_tf(boxes1, boxes2):
b1_len = boxes1.shape[0]
b2_len = boxes2.shape[0]
boxes1_0 = tf.reshape(boxes1[:, 0],[b1_len,1])
boxes1_1 = tf.reshape(boxes1[:, 1],[b1_len,1])
boxes1_2 = tf.reshape(boxes1[:, 2],[b1_len,1])
boxes1_3 = tf.reshape(boxes1[:, 3],[b1_len,1])
boxes2_0 = tf.reshape(boxes2[:, 0],[1,b2_len])
boxes2_1 = tf.reshape(boxes2[:, 1],[1,b2_len])
boxes2_2 = tf.reshape(boxes2[:, 2],[1,b2_len])
boxes2_3 = tf.reshape(boxes2[:, 3],[1,b2_len])
dx_left = tf.maximum(boxes1_0 - boxes1_2 / 2, boxes2_0 - boxes2_2 / 2)
dx_right = tf.minimum(boxes1_0 + boxes1_2 / 2, boxes2_0 + boxes2_2 / 2)
dx_overlap = dx_right - dx_left
dx_overlap = tf.maximum(dx_overlap, tf.zeros_like(dx_overlap))
dy_left = tf.maximum(boxes1_1 - boxes1_3 / 2, boxes2_1 - boxes2_3 / 2)
dy_right = tf.minimum(boxes1_1 + boxes1_3 / 2, boxes2_1 + boxes2_3 / 2)
dy_overlap = dy_right - dy_left
dy_overlap = tf.maximum(dy_overlap, tf.zeros_like(dy_overlap))
intersections = dx_overlap * dy_overlap
unions = boxes1_2 * boxes1_3 + boxes2_2 * boxes2_3 - intersections
return intersections / tf.maximum(unions,1e-10)
def build_targets_masks(pred_boxes, ground_truth, batch_size, anchors_num, nH, nW, thresh, reduction):
orgimg_w = 304
orgimg_h = 224
nAnchors = anchors_num*nH*nW
object_no_detections = []
batch_num = 32
for b in range(batch_num):
ground_truth_tensor = tf.reshape(ground_truth[b],[-1,5])
groundtruth_num = tf.shape(ground_truth_tensor)[0]
# if tf.reduce_sum(ground_truth_tensor) == 0: # No gt for this image
# print("tf.reduce_sum(ground_truth_tensor) == 0")
# continue
cur_pred_boxes = pred_boxes[b*nAnchors:(b+1)*nAnchors]
scale = tf.constant([[orgimg_w,orgimg_h,orgimg_w,orgimg_h]],tf.float32)
gt = ground_truth_tensor[:,1:5] * scale / tf.constant(reduction)
# Set confidence mask of matching detections to 0
iou_gt_pred = bbox_ious_tf(gt, cur_pred_boxes)
iou_gt_pred = tf.reshape(iou_gt_pred,[1,groundtruth_num,anchors_num,nH*nW])
mask = tf.cast((tf.reduce_sum(tf.cast((iou_gt_pred > thresh),tf.float32),1) >= 1),tf.float32)
object_no_mask = tf.ones_like(mask, tf.float32) - mask
if object_no_detections==[]:
object_no_detections = object_no_mask
else:
object_no_detections = tf.concat([object_no_detections,object_no_mask],0)
return object_no_detections
def yolo_loss( target,
output,
coord_mask, object_detections, object_no_detections_gt_anch, gt_coord, gt_conf,
object_scale=5.0,
no_object_scale=1.0, coordinates_scale=3.0
):
#共有5个anchor，16为网路输入到输出的下采样倍率，由于最后的iou计算以输出feature map为基准
#所以除以16，五个anchor等长宽分别为1,3,5,7,10
anchors = tf.constant([[16, 16], [48, 48], [80, 80], [112,112], [160,160]], tf.float32) / tf.constant(16.0)
anchors_num = tf.shape(anchors)[0]
output_shape = tf.shape(output) #(n,c,h,w)
batch_size = output_shape[0]
nH = output_shape[2]
nW = output_shape[3]
#reshape output as [batch,anchor,x-y-w-h-conf,w*h]
output = tf.reshape(output,[batch_size, anchors_num, -1, nW*nH])
#将output的x、y归一化到0-1
coord_xy = tf.sigmoid(output[:, :, :2]) # x,y
coord_wh = output[:, :, 2:4] # w,h
pre_coord = tf.concat([coord_xy,coord_wh], 2)
#将置信度归一化到0-1
pre_conf = tf.sigmoid(output[:, :, 4])
# Create prediction boxes
lin_x = tf.reshape(tf.tile(tf.reshape(tf.linspace(0.0, tf.cast(nW - 1,tf.float32), nW),[1,nW]),[nH,1]),[nW*nH])
lin_y = tf.reshape(tf.tile(tf.reshape(tf.linspace(0.0, tf.cast(nH - 1,tf.float32), nH),[nH,1]),[1,nW]),[nW*nH])
#将归一化后的x、y（相当于bias）分别加上对应所在网格的x、y整数值
pred_boxes_x = tf.reshape(pre_coord[:, :, 0] + lin_x,[-1,1])
pred_boxes_y = tf.reshape(pre_coord[:, :, 1] + lin_y,[-1,1])
#解码过程 对应于后面的编码过程
anchor_w = tf.reshape(anchors[:, 0],[anchors_num,1])
anchor_h = tf.reshape(anchors[:, 1],[anchors_num,1])
pred_boxes_w = tf.reshape(tf.exp(pre_coord[:, :, 2]) * anchor_w,[-1,1])
pred_boxes_h = tf.reshape(tf.exp(pre_coord[:, :, 3]) * anchor_h,[-1,1])
#x,y,w,h
#将predict boxes reshape成anchors*x*y行4列的tensor（4列指x,y,w,h）
pred_boxes = tf.concat([pred_boxes_x,pred_boxes_y,pred_boxes_w,pred_boxes_h],1)
#根据predict box及groundtruth box计算iou获得一定没有物体所在的网格（每个网格点代表一个box）
object_no_detections_gt_pre = build_targets_masks(pred_boxes, target, batch_size, anchors_num, nH, nW, thresh = 0.6, reduction = 16.0)
#object_no_detections_gt_anch为传入参数，由gt所在的best anchor位置的x,y确定，乘以
#object_no_detections_gt_pre后将object_no_detections_gt_anch为0的位置强制置零，即gt
#的位置no_object一定为0
object_no_detections = object_no_detections_gt_pre * object_no_detections_gt_anch
# coord
#coord_mask为2-gt_w*gt_h/w*h,为小框加大系数，使小框更容易检测到，坐标相关的loss都要乘
coord_mask = tf.tile(coord_mask,[1,1,2,1])
pre_coord_center, gt_coord_center = pre_coord[:, :, :2], gt_coord[:, :, :2]
pre_coord_wh, gt_coord_wh = pre_coord[:, :, 2:], gt_coord[:, :, 2:]
# Compute losses
#中心点loss使用二值交叉熵loss，宽高loss采用smooth l1 loss
loss_coord_center = 2.0 * 1.0 * coordinates_scale * tf.reduce_sum(coord_mask * tf.keras.backend.binary_crossentropy(gt_coord_center, pre_coord_center))
loss_coord_wh = 2.0 * 1.5 * coordinates_scale * tf.reduce_sum(coord_mask * tf.losses.huber_loss(gt_coord_wh, pre_coord_wh, reduction=tf.losses.Reduction.NONE))
loss_coord = loss_coord_center + loss_coord_wh
#正负样本置信度均采用二值交叉熵loss
loss_conf_pos = 1.0 * object_scale * tf.reduce_sum(object_detections * tf.keras.backend.binary_crossentropy(gt_conf, pre_conf))
loss_conf_neg = 1.0 * no_object_scale * tf.reduce_sum(object_no_detections * tf.keras.backend.binary_crossentropy(gt_conf, pre_conf))
loss_conf = loss_conf_pos + loss_conf_neg
loss_tot = (loss_coord + loss_conf) / tf.cast(batch_size,tf.float32)
return loss_tot