#----------------------------------------------------#
if __name__ == "__main__":
config = Config()
#----------------------------------------------------#
# 训练之前一定要修改NUM_CLASSES
# 修改成所需要区分的类的个数+1。
#----------------------------------------------------#
NUM_CLASSES = 21
#-----------------------------------------------------#
# input_shape是输入图片的大小，默认为800,800,3
# 随着输入图片的增大，占用显存会增大
# 视频上为600,600,3，多次训练测试后发现800,800,3更优
#-----------------------------------------------------#
input_shape = [800, 800, 3]
model_rpn, model_all = get_model(config, NUM_CLASSES)
#------------------------------------------------------#
# 权值文件请看README，百度网盘下载
# 训练自己的数据集时提示维度不匹配正常
# 预测的东西都不一样了自然维度不匹配
#------------------------------------------------------#
base_net_weights = "model_data/voc_weights.h5"
model_rpn.load_weights(base_net_weights, by_name=True)
model_all.load_weights(base_net_weights, by_name=True)
bbox_util = BBoxUtility(overlap_threshold=config.rpn_max_overlap,
ignore_threshold=config.rpn_min_overlap,
top_k=config.num_RPN_train_pre)
#--------------------------------------------#
# 训练参数的设置
config = Config()
NUM_CLASSES = 21
# 训练100世代
EPOCH = 100
# 开始使用1e-4训练，每过10个世代降低为原来的1/2
Learning_rate = 1e-4
bbox_util = BBoxUtility(overlap_threshold=config.rpn_max_overlap,
ignore_threshold=config.rpn_min_overlap)
annotation_path = '2007_train.txt'
#------------------------------------------------------#
# 权值文件请看README，百度网盘下载
# 训练自己的数据集时提示维度不匹配正常
# 预测的东西都不一样了自然维度不匹配
#------------------------------------------------------#
model_rpn, model_classifier, model_all = get_model(config, NUM_CLASSES)
base_net_weights = "model_data/voc_weights.h5"
model_all.summary()
model_rpn.load_weights(base_net_weights, by_name=True)
model_classifier.load_weights(base_net_weights, by_name=True)
with open(annotation_path) as f:
lines = f.readlines()
np.random.seed(10101)
np.random.shuffle(lines)
np.random.seed(None)
# 每个世代训练数据集长度的步数
# 根据数据集大小进行指定
EPOCH_LENGTH = len(lines)
#--------------------------------------------#
# 该部分代码用于看网络结构
#--------------------------------------------#
from nets.frcnn import get_model
from utils.utils import net_flops
if __name__ == "__main__":
input_shape = [600, 600]
num_classes = 21
_, model = get_model(num_classes, 'vgg', input_shape=[input_shape[0], input_shape[1], 3])
#--------------------------------------------------------#
# 由于faster-rcnn-keras里面存在一些不可计算层
# 如ROIpooling
# 因此无法正确计算FLOPs，可参考pytorch版本
#--------------------------------------------------------#
#--------------------------------------------#
# 查看网络结构网络结构
#--------------------------------------------#
model.summary()
#--------------------------------------------#
# 计算网络的FLOPS
#--------------------------------------------#
net_flops(model, table=False)
#--------------------------------------------#
# 获得网络每个层的名称与序号
#--------------------------------------------#
# for i,layer in enumerate(model.layers):
# print(i,layer.name)
#--------------------------------------------#
# 该部分代码用于看网络结构
#--------------------------------------------#
from nets.frcnn import get_model
if __name__ == "__main__":
num_classes = 21
_, model = get_model(num_classes, 'resnet50')
model.summary()
for i, layer in enumerate(model.layers):
print(i, layer.name)
Freeze_Train = True
#----------------------------------------------------#
# 获得图片路径和标签
#----------------------------------------------------#
train_annotation_path = '2007_train.txt'
val_annotation_path = '2007_val.txt'
#----------------------------------------------------#
# 获取classes和anchor
#----------------------------------------------------#
class_names, num_classes = get_classes(classes_path)
num_classes += 1
anchors = get_anchors(input_shape, backbone, anchors_size)
K.clear_session()
model_rpn, model_all = get_model(num_classes, backbone=backbone)
if model_path != '':
#------------------------------------------------------#
# 载入预训练权重
#------------------------------------------------------#
print('Load weights {}.'.format(model_path))
model_rpn.load_weights(model_path, by_name=True)
model_all.load_weights(model_path, by_name=True)
#--------------------------------------------#
# 训练参数的设置
#--------------------------------------------#
callback = tf.summary.create_file_writer("logs")
loss_history = LossHistory("logs/")
bbox_util = BBoxUtility(num_classes)
#------------------------------------------------------#
# 设置用到的显卡
#------------------------------------------------------#
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(str(x) for x in train_gpu)
ngpus_per_node = len(train_gpu)
print('Number of devices: {}'.format(ngpus_per_node))
#----------------------------------------------------#
# 获取classes和anchor
#----------------------------------------------------#
class_names, num_classes = get_classes(classes_path)
num_classes += 1
anchors = get_anchors(input_shape, backbone, anchors_size)
K.clear_session()
model_rpn_body, model_all_body = get_model(num_classes, backbone=backbone)
if model_path != '':
#------------------------------------------------------#
# 载入预训练权重
#------------------------------------------------------#
print('Load weights {}.'.format(model_path))
model_rpn_body.load_weights(model_path, by_name=True)
model_all_body.load_weights(model_path, by_name=True)
if ngpus_per_node > 1:
model_rpn = multi_gpu_model(model_rpn_body, gpus=ngpus_per_node)
model_all = multi_gpu_model(model_all_body, gpus=ngpus_per_node)
else:
model_rpn = model_rpn_body
model_all = model_all_body
