Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
ssd_training.py 5.35 KB
Newer Older
zhenyi's avatar
tf2 detection  
zhenyi committed
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 
import tensorflow as tf


class MultiboxLoss(object):
    def __init__(self, num_classes, alpha=1.0, neg_pos_ratio=3.0,
                 background_label_id=0, negatives_for_hard=100.0):
        self.num_classes = num_classes
        self.alpha = alpha
        self.neg_pos_ratio = neg_pos_ratio
        if background_label_id != 0:
            raise Exception('Only 0 as background label id is supported')
        self.background_label_id = background_label_id
        self.negatives_for_hard = negatives_for_hard

    def _l1_smooth_loss(self, y_true, y_pred):
        abs_loss = tf.abs(y_true - y_pred)
        sq_loss = 0.5 * (y_true - y_pred)**2
        l1_loss = tf.where(tf.less(abs_loss, 1.0), sq_loss, abs_loss - 0.5)
        return tf.reduce_sum(l1_loss, -1)

    def _softmax_loss(self, y_true, y_pred):
        y_pred = tf.maximum(y_pred, 1e-7)
        softmax_loss = -tf.reduce_sum(y_true * tf.math.log(y_pred),
                                      axis=-1)
        return softmax_loss

    def compute_loss(self, y_true, y_pred):
        # --------------------------------------------- #
        #   y_true batch_size, 8732, 4 + self.num_classes + 1
        #   y_pred batch_size, 8732, 4 + self.num_classes
        # --------------------------------------------- #
        num_boxes = tf.cast(tf.shape(y_true)[1], tf.float32)

        # --------------------------------------------- #
        #   分类的loss
        #   batch_size,8732,21 -> batch_size,8732
        # --------------------------------------------- #
        conf_loss = self._softmax_loss(y_true[:, :, 4:-1],
                                       y_pred[:, :, 4:])
        # --------------------------------------------- #
        #   框的位置的loss
        #   batch_size,8732,4 -> batch_size,8732
        # --------------------------------------------- #
        loc_loss = self._l1_smooth_loss(y_true[:, :, :4],
                                        y_pred[:, :, :4])

        # --------------------------------------------- #
        #   获取所有的正标签的loss
        # --------------------------------------------- #
        pos_loc_loss = tf.reduce_sum(loc_loss * y_true[:, :, -1],
                                     axis=1)
        pos_conf_loss = tf.reduce_sum(conf_loss * y_true[:, :, -1],
                                      axis=1)

        # --------------------------------------------- #
        #   每一张图的正样本的个数
        #   num_pos     [batch_size,]
        # --------------------------------------------- #
        num_pos = tf.reduce_sum(y_true[:, :, -1], axis=-1)

        # --------------------------------------------- #
        #   每一张图的负样本的个数
        #   num_neg     [batch_size,]
        # --------------------------------------------- #
        num_neg = tf.minimum(self.neg_pos_ratio * num_pos, num_boxes - num_pos)
        # 找到了哪些值是大于0的
        pos_num_neg_mask = tf.greater(num_neg, 0)
        # --------------------------------------------- #
        #   如果所有的图，正样本的数量均为0
        #   那么则默认选取100个先验框作为负样本
        # --------------------------------------------- #
        has_min = tf.cast(tf.reduce_any(pos_num_neg_mask), tf.float32)
        num_neg = tf.concat(axis=0, values=[num_neg, [(1 - has_min) * self.negatives_for_hard]])
        
        # --------------------------------------------- #
        #   从这里往后，与视频中看到的代码有些许不同。
        #   由于以前的负样本选取方式存在一些问题，
        #   我对该部分代码进行重构。
        #   求整个batch应该的负样本数量总和
        # --------------------------------------------- #
        num_neg_batch = tf.reduce_sum(tf.boolean_mask(num_neg, tf.greater(num_neg, 0)))
        num_neg_batch = tf.cast(num_neg_batch, tf.int32)

        # --------------------------------------------- #
        #   对预测结果进行判断，如果该先验框没有包含物体
        #   那么它的不属于背景的预测概率过大的话
        #   就是难分类样本
        # --------------------------------------------- #
        confs_start = 4 + self.background_label_id + 1
        confs_end   = confs_start + self.num_classes - 1

        # --------------------------------------------- #
        #   batch_size,8732
        #   把不是背景的概率求和，求和后的概率越大
        #   代表越难分类。
        # --------------------------------------------- #
        max_confs = tf.reduce_sum(y_pred[:, :, confs_start:confs_end], axis=2)

        # --------------------------------------------------- #
        #   只有没有包含物体的先验框才得到保留
        #   我们在整个batch里面选取最难分类的num_neg_batch个
        #   先验框作为负样本。
        # --------------------------------------------------- #
        max_confs   = tf.reshape(max_confs * (1 - y_true[:, :, -1]), [-1])
        _, indices  = tf.nn.top_k(max_confs, k=num_neg_batch)

        neg_conf_loss = tf.gather(tf.reshape(conf_loss, [-1]), indices)

        # 进行归一化
        num_pos     = tf.where(tf.not_equal(num_pos, 0), num_pos, tf.ones_like(num_pos))
        total_loss  = tf.reduce_sum(pos_conf_loss) + tf.reduce_sum(neg_conf_loss) + tf.reduce_sum(self.alpha * pos_loc_loss)
        total_loss /= tf.reduce_sum(num_pos)
        return total_loss