资源简介
3dcnn
行为识别网络架构并使用softmax层
用于ucf101数据集
代码片段和文件信息
import tensorflow as tf
import vw_c3d_tools as c3d_tools
N_CLASSE = 9
def C3D_MODEL(xn_classesis_pretrain=True):
# conv(layer_name x out_channels weight_decay biases_decay kernel_size=[333] stride=[11111] is_pretrain=True):
# pool(layer_name x k is_max_pool=True):
# FC_layer(layer_name x out_nodesweight_decay biases_decay):
parameters = []
conv1 = c3d_tools.conv(“conv1“ x 64 0.0005 0 is_pretrain=is_pretrain)
with tf.name_scope(‘pool1‘):
pool1 = c3d_tools.pool(“pool1“ conv1 1 is_max_pool=True)
conv2 = c3d_tools.conv(“conv2“ pool1 128 0.0005 0 is_pretrain=is_pretrain)
with tf.name_scope(‘pool2‘):
pool2 = c3d_tools.pool(“pool2“ conv2 2 is_max_pool=True)
conv3a = c3d_tools.conv(“conv3a“ pool2 256 0.0005 0 is_pretrain=is_pretrain)
conv3b = c3d_tools.conv(“conv3b“ conv3a 256 0.0005 0 is_pretrain=is_pretrain)
with tf.name_scope(‘pool3‘):
pool3 = c3d_tools.pool(“pool3“ conv3b 2 is_max_pool=True)
conv4a = c3d_tools.conv(“conv4a“ pool3 512 0.0005 0 is_pretrain=is_pretrain)
conv4b = c3d_tools.conv(“conv4b“ conv4a 512 0.0005 0 is_pretrain=is_pretrain)
with tf.name_scope(‘pool4‘):
pool4 = c3d_tools.pool(“pool4“ conv4b 2 is_max_pool=True)
conv5a = c3d_tools.conv(“conv5a“ pool4 512 0.0005 0 is_pretrain=is_pretrain)
conv5b = c3d_tools.conv(“conv5b“ conv5a 512 0.0005 0 is_pretrain=is_pretrain)
with tf.name_scope(‘pool5‘):
pool5 = c3d_tools.pool(“pool5“ conv5b 2 is_max_pool=True)
pool5 = tf.transpose(pool5 perm=[01423])
with tf.name_scope(‘fc6‘):
fc6 = c3d_tools.FC_layer(‘fc6‘ pool5 4096 0.0005 0 0.5)
fc6 = c3d_tools.batch_norm(fc6)
with tf.name_scope(‘fc7‘):
fc7 = c3d_tools.FC_layer(‘fc7‘ fc6 4096 0.0005 0 0.5)
fc7 = c3d_tools.batch_norm(fc7)
with tf.name_scope(‘fc8‘):
fc8 = c3d_tools.FC_layer(‘fc8‘ fc7 n_classes 0.0005 0 1)
fc8= c3d_tools.batch_norm(fc8)
return fc8
‘‘‘with tf.name_scope(“softmax“):
softmax = tf.nn.softmax(tf.matmul(fc8W) + b)
return softmax‘‘‘
with tf.name_scope(‘softmax1‘) as scope:
w = tf.Variable(tf.truncated_normal([256N_CLASSE]
dtype=tf.float32
stddev=1e-1)name=‘weights‘)
b = tf.Variable(tf.constant(0.0 shape=[N_CLASSE] dtype=tf.float32)
- 上一篇:快来一起挖掘幸福感吧!.zip
- 下一篇:能够标注人脸边界框和五个关键点
相关资源
- lstm_tensorflow
- 《TensorFlow2深度学习》
- TensorFlow Python API documentation.pdf
- Python-使用最新版本的tensorflow实现se
- BP神经网络及代码分析(python+tensorf
- Python-用TensorFlow实现神经网络实体关系
- 莫烦全部代码Reinforcement-learning-with-
- Python-基于TensorFlow和BERT的管道式实体
- Tensorflow gpu_accelerate
- python tensorFlow AND和XOR
- word2vec.py(来自黄文坚的“tensorflow实
- keras+tensorflow CNN
- 基于机器学习框架tensorflow的图像分类
- 机器学习实战:基于 Scikit-Learn 和 T
- 《白话深度学习与TensorFlow》.pdf
- Tensorflow+实战Google深度学习框架
- python MNIST分类 tensorflow
- tensorflow下用LSTM网络进行时间序列预测
- 深度学习之二:用Tensorflow实现卷积神
- tensorflow版本的YOLO v3,在Windows系统下
- Python实现的TensorFlow入门案例
- 《TensorFlow技术解析与实战》高清中文
- python实现TensorFlow2 股票股价预测(源
- 计算 语义分割结果 MIOU,miou.py操作简
- tensorflow做cifar-10识别
- TensorFlow手写数字识别代码
- 基于kNN方法的MNIST手写数字识别Tenso
- cifar10_input.py
- Python-PythonTensorflowKeras实现参数tSNE算法
- TensorFlow实现CNN
评论
共有 条评论