資源簡介
3dcnn
行為識別網絡架構并使用softmax層
用于ucf101數據集
代碼片段和文件信息
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:快來一起挖掘幸福感吧!.zip
- 下一篇:能夠標注人臉邊界框和五個關鍵點
評論
共有 條評論