def get_classifier(base_layers,
input_rois,
num_rois,
nb_classes=21,
trainable=False):
''' roi --> cls+reg'''
pooling_regions = 14
input_shape = (num_rois, 14, 14, 1024)
# base_layers[38,38,1024], input_rois[num_prior,4] num_prior=32
# 1 roiPooling，base_layers[38,38,1024], input_rois[-,4] ,pooling_regions=14, num_rois=32
out_roi_pool = RoiPoolingConv(pooling_regions,
num_rois)([base_layers,
input_rois]) # input_rois是建议框
# 2
out = classifier_layers(out_roi_pool,
input_shape=input_shape,
trainable=True)
out = TimeDistributed(Flatten())(out)
out_class = TimeDistributed(Dense(nb_classes,
activation='softmax',
kernel_initializer='zero'),
name='dense_class_{}'.format(nb_classes))(out)
out_regr = TimeDistributed(Dense(4 * (nb_classes - 1),
activation='linear',
kernel_initializer='zero'),
name='dense_regress_{}'.format(nb_classes))(out)
return [out_class, out_regr] # 21, 20*4
def classifier(self, base_layers, trainable=True):
out_roi_pool = RoiPoolingConv(
self.pooling_region, self.num_rois)([base_layers, self.roi_input])
out = TimeDistributed(Flatten(name='flatten'))(out_roi_pool)
out = TimeDistributed(
Dense(4096, activation='relu', name='fc1',
trainable=trainable))(out)
out = TimeDistributed(Dropout(0.5))(out)
out = TimeDistributed(
Dense(4096, activation='relu', name='fc2',
trainable=trainable))(out)
out = TimeDistributed(Dropout(0.5))(out)
out_class = TimeDistributed(Dense(self.num_cls,
activation='softmax',
kernel_initializer='zero',
trainable=trainable),
name='dense_class_{}'.format(
self.num_cls))(out)
# note: no regression target for bg class
out_regr = TimeDistributed(Dense(4 * (self.num_cls - 1),
activation='linear',
kernel_initializer='zero',
trainable=trainable),
name='dense_regress_{}'.format(
self.num_cls))(out)
return [out_class, out_regr]
def get_classifier(base_layers,
input_rois,
num_rois,
nb_classes=21,
trainable=False):
# 设定池化大小
pooling_regions = 14
# 设定输入大小（roi数量，池化大小，池化大小，固定长度）
input_shape = (num_rois, 14, 14, 1024)
# base_layers -> [38,38,1024] 当输入为[600,600,3] 的图片的时候
#input_rois -> [None, 4]，None一般每次处理为32个建议框，在config.py中的num.rois中设置，
# 获取RoiPoolingConv层输出的公共特征层
out_roi_pool = RoiPoolingConv(pooling_regions,
num_rois)([base_layers, input_rois])
# 获取分类层的输出结果
out = classifier_layers(out_roi_pool,
input_shape=input_shape,
trainable=True)
# 进行flatten操作，准备进行全连接输出分类结果
out = TimeDistributed(Flatten())(out)
# 进行全连接输出分类结果
out_class = TimeDistributed(Dense(nb_classes,
activation="softmax",
kernel_initializer="zero"),
name="dense_class_{}".format(nb_classes))(out)
# 进行全连接输出调整后的框位置
out_regr = TimeDistributed(Dense(4 * (nb_classes - 1),
activation="linear",
kernel_initializer="zero"),
name="dense_regress_{}".format(nb_classes))(out)
# 返回
return [out_class, out_regr]
def get_ROIout(base_layers,
input_rois,
num_rois,
nb_classes=21,
trainable=False): #input_rois （None，4）若干个建议框
pooling_regions = 14
out_roi_pool = RoiPoolingConv(pooling_regions, num_rois)(
[base_layers, input_rois]) #输入featuremap和建议框 返回num_rois个14*14的roi
return out_roi_pool
def get_classifier(base_layers, input_rois, nb_classes=21, pooling_regions=14):
# num_rois, 38, 38, 1024 -> num_rois, 14, 14, 2048
out_roi_pool = RoiPoolingConv(pooling_regions)([base_layers, input_rois])
# num_rois, 14, 14, 1024 -> num_rois, 1, 1, 2048
out = classifier_layers(out_roi_pool)
# num_rois, 1, 1, 1024 -> num_rois, 2048
out = TimeDistributed(Flatten())(out)
# num_rois, 1, 1, 1024 -> num_rois, nb_classes
out_class = TimeDistributed(Dense(nb_classes,
activation='softmax',
kernel_initializer='zero'),
name='dense_class_{}'.format(nb_classes))(out)
# num_rois, 1, 1, 1024 -> num_rois, 4 * (nb_classes-1)
out_regr = TimeDistributed(Dense(4 * (nb_classes - 1),
activation='linear',
kernel_initializer='zero'),
name='dense_regress_{}'.format(nb_classes))(out)
return [out_class, out_regr]
def get_classifier(base_layers,
input_rois,
num_rois,
nb_classes=21,
trainable=False):
pooling_regions = 14
input_shape = (num_rois, 14, 14, 1024)
out_roi_pool = RoiPoolingConv(pooling_regions,
num_rois)([base_layers, input_rois])
out = classifier_layers(out_roi_pool,
input_shape=input_shape,
trainable=True)
out = TimeDistributed(Flatten())(out)
out_class = TimeDistributed(Dense(nb_classes,
activation='softmax',
kernel_initializer='zero'),
name='dense_class_{}'.format(nb_classes))(out)
out_regr = TimeDistributed(Dense(4 * (nb_classes - 1),
activation='linear',
kernel_initializer='zero'),
name='dense_regress_{}'.format(nb_classes))(out)
return [out_class, out_regr]
def get_classifier(base_layers, input_rois, nb_classes=21, pooling_regions=14):
'''
Faster-RCNN网络模型
Parameters
----------
base_layers: resnet50输出的特征层（None,38,38,1024）
input_rois:
nb_classes
pooling_regions
Returns
-------
'''
# num_rois:一张图片中建议框数量
# num_rois, 38, 38, 1024 -> num_rois, 14, 14, 2048
out_roi_pool = RoiPoolingConv()([base_layers, input_rois, pooling_regions])
# out_roi_pool = RoiPoolingConv(pooling_regions)([base_layers, input_rois])
# num_rois, 14, 14, 1024 -> num_rois, 1, 1, 2048
out = classifier_layers(out_roi_pool)
# TimeDistributed: 对batch_size中的每一个单独做处理
# num_rois, 1, 1, 1024 -> num_rois, 2048
out = TimeDistributed(Flatten())(out)
# num_rois, 1, 1, 1024 -> num_rois, nb_classes
# (batch_size,num_rois,nb_classes),None:batch_size,num_rois:每张图片的建议框数量,nb_classes:每个建议框预测的类别
proposal_boxes_class = TimeDistributed(
Dense(nb_classes, activation='softmax', kernel_initializer='zero'),
name=f'dense_class_{nb_classes}')(out)
# num_rois, 1, 1, 1024 -> num_rois, 4 * (nb_classes-1)
# (batch_size,num_rois,4*(nb_classes-1)), 4*(nb_classes-1):每个建议框预测的所有类别的建议框变化量. 这个变化量+建议框=预测框
proposal_boxes_offset = TimeDistributed(
Dense(4 * (nb_classes - 1),
activation='linear',
kernel_initializer='zero'),
name=f'dense_regress_{nb_classes}')(out)
return [proposal_boxes_class, proposal_boxes_offset]