KnightZhang625 / Enhanced-RCNN_TensorFlow

TensorFlow implementation for the Enhanced-RCNN
Apache License 2.0
4 stars 5 forks source link

你好?请问可以告诉您代码里面关于CNN卷积的细节处理? #5

Open nutshell999 opened 3 years ago

nutshell999 commented 3 years ago

如果可以的话,我想知道CNN的输入规格以及CNN卷积时卷积核是在哪个维度进行移动?麻烦您了

nutshell999 commented 3 years ago

`

W = tf.Variable(tf.truncated_normal(filter_shape, stddev=0.1), name='W') b = tf.Variable(tf.truncated_normal([1], stddev=0.1), name='b') conv = tf.nn.conv2d(inputs, W, strides=[1, 1, 1, 1], padding='VALID', name='conv') h = tf.nn.relu(tf.nn.bias_add(conv, b), name='relu')

`

nutshell999 commented 3 years ago

貌似卷积核不是一个整数型的卷积核,是一个随机正态截取的浮点型卷积核吗?