fasterrcnn论文_机器学习-Faster RCNN的损失函数(Loss Function)

Faster RCNN的损失函数(Loss Function)的形式如下:

p(i): Anchor[i]的预测分类概率；

Anchor[i]是正样本时,p(i)*=1；Anchor[i]是负样本时，p(i)*=0;

什么是正样本与负样本 满足以下条件的Anchor是正样本：与Ground Truth Box的IOU(Intersection-Over-Union) 的重叠区域最大的Anchor；与Gound Truth Box的IOU的重叠区域>0.7; 满足以下条件的Anchor是负样本：与Gound Truth Box的IOU的重叠区域 <0.3;既不属于正样本又不属于负样本的Anchor不参与训练。

t(i): Anchor[i]预测的Bounding Box的参数化坐标(parameterized coordinates)；

t(i)*: Anchor[i]的Ground Truth的Bounding Box的参数化坐标；

N(cls): mini-batch size;

N(reg): Anchor Location的数量;

其中，R是Smooth L1函数；

表示只有在正样本时才回归Bounding Box。

Smooth L1 Loss

Smooth L1完美地避开了 L1 和 L2 损失的缺陷，在 x 较小时，对 x 的梯度也会变小; 而在 x 很大时，对 x 的梯度的绝对值达到上限1，不会因预测值的梯度十分大导致训练不稳定。

L(cls): 是两个类别的对数损失

λ: 权重平衡参数，在论文中作者设置λ=10，但实际实验显示，结果对的λ变化不敏感，如下表所示，λ取值从1变化到100，对最终结果的影响在1%以内。

代码实现

Smooth L1 Loss

def _smooth_l1_loss(self, bbox_pred, bbox_targets, bbox_inside_weights, bbox_outside_weights, sigma=1.0, dim=[1]):

sigma_2 = sigma ** 2

box_diff = bbox_pred - bbox_targets

in_box_diff = bbox_inside_weights * box_diff

abs_in_box_diff = tf.abs(in_box_diff)

smoothL1_sign = tf.stop_gradient(tf.to_float(tf.less(abs_in_box_diff, 1. / sigma_2)))

in_loss_box = tf.pow(in_box_diff, 2) * (sigma_2 / 2.) * smoothL1_sign

+ (abs_in_box_diff - (0.5 / sigma_2)) * (1. - smoothL1_sign)

out_loss_box = bbox_outside_weights * in_loss_box

loss_box = tf.reduce_mean(tf.reduce_sum(

out_loss_box,

axis=dim

))

return loss_box

代码中的Smooth L1 Loss更加General。

bbox_inside_weight对应于公式(1)(Faster RCNN的损失函数)中的p*，即当Anchor为正样本时值为1，为负样本时值为0。bbox_outside_weights对应于公式(1)(Faster RCNN的损失函数)中的N(reg)、λ、N(cls)的设置。在论文中，N(reg)=2400、λ=10、N(cls)=256，如此分类和回归两个loss的权重基本相同。

在代码中，N(reg)=N(cls),λ=1，如此分类和回归两个loss的权重也基本相同。

Loss

def _add_losses(self, sigma_rpn=3.0):

with tf.variable_scope('LOSS_' + self._tag) as scope:

# RPN, class loss

rpn_cls_score = tf.reshape(self._predictions['rpn_cls_score_reshape'], [-1, 2])

rpn_label = tf.reshape(self._anchor_targets['rpn_labels'], [-1])

rpn_select = tf.where(tf.not_equal(rpn_label, -1))

rpn_cls_score = tf.reshape(tf.gather(rpn_cls_score, rpn_select), [-1, 2])

rpn_label = tf.reshape(tf.gather(rpn_label, rpn_select), [-1])

rpn_cross_entropy = tf.reduce_mean(

tf.nn.sparse_softmax_cross_entropy_with_logits(logits=rpn_cls_score, labels=rpn_label))

# RPN, bbox loss

rpn_bbox_pred = self._predictions['rpn_bbox_pred']

rpn_bbox_targets = self._anchor_targets['rpn_bbox_targets']

rpn_bbox_inside_weights = self._anchor_targets['rpn_bbox_inside_weights']

rpn_bbox_outside_weights = self._anchor_targets['rpn_bbox_outside_weights']

rpn_loss_box = self._smooth_l1_loss(rpn_bbox_pred, rpn_bbox_targets, rpn_bbox_inside_weights,

rpn_bbox_outside_weights, sigma=sigma_rpn, dim=[1, 2, 3])

# RCNN, class loss

cls_score = self._predictions["cls_score"]

label = tf.reshape(self._proposal_targets["labels"], [-1])

cross_entropy = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=cls_score, labels=label))

# RCNN, bbox loss

bbox_pred = self._predictions['bbox_pred']

bbox_targets = self._proposal_targets['bbox_targets']

bbox_inside_weights = self._proposal_targets['bbox_inside_weights']

bbox_outside_weights = self._proposal_targets['bbox_outside_weights']

loss_box = self._smooth_l1_loss(bbox_pred, bbox_targets, bbox_inside_weights, bbox_outside_weights)

self._losses['cross_entropy'] = cross_entropy

self._losses['loss_box'] = loss_box

self._losses['rpn_cross_entropy'] = rpn_cross_entropy

self._losses['rpn_loss_box'] = rpn_loss_box

loss = cross_entropy + loss_box + rpn_cross_entropy + rpn_loss_box

regularization_loss = tf.add_n(tf.losses.get_regularization_losses(), 'regu')

self._losses['total_loss'] = loss + regularization_loss

self._event_summaries.update(self._losses)

return loss

损失函数中包含了RPN交叉熵、RPN Box的regression、RCNN的交叉熵、RCNN Box的regression以及参数正则化损失。

IOU的计算

def bbox_overlaps(

np.ndarray[DTYPE_t, ndim=2] boxes,

np.ndarray[DTYPE_t, ndim=2] query_boxes):

"""

Parameters

----------

boxes: (N, 4) ndarray of float

query_boxes: (K, 4) ndarray of float

Returns

-------

overlaps: (N, K) ndarray of overlap between boxes and query_boxes

"""

cdef unsigned int N = boxes.shape[0]

cdef unsigned int K = query_boxes.shape[0]

cdef np.ndarray[DTYPE_t, ndim=2] overlaps = np.zeros((N, K), dtype=DTYPE)

cdef DTYPE_t iw, ih, box_area

cdef DTYPE_t ua

cdef unsigned int k, n

for k in range(K):

box_area = (

(query_boxes[k, 2] - query_boxes[k, 0] + 1) *

(query_boxes[k, 3] - query_boxes[k, 1] + 1)

)

for n in range(N):

iw = (

min(boxes[n, 2], query_boxes[k, 2]) -

max(boxes[n, 0], query_boxes[k, 0]) + 1

)

if iw > 0:

ih = (

min(boxes[n, 3], query_boxes[k, 3]) -

max(boxes[n, 1], query_boxes[k, 1]) + 1

)

if ih > 0:

ua = float(

(boxes[n, 2] - boxes[n, 0] + 1) *

(boxes[n, 3] - boxes[n, 1] + 1) +

box_area - iw * ih

)

overlaps[n, k] = iw * ih / ua

return overlaps

IOU覆盖率的计算过程：IOU=C/(A+B-C)

IOU计算