Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- image_size = 128
- num_channels = 1
- batch = 128
- descriptor_size = 512
- def weight_variable(shape):
- initial = tf.truncated_normal(shape, stddev=0.1)
- return tf.Variable(initial)
- def bias_variable(shape):
- initial = tf.constant(0.1, shape=shape)
- return tf.Variable(initial)
- def conv2d(x, W):
- return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
- def max_pool_2x2(x):
- return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
- #INPUT
- xs = tf.placeholder(tf.float32, [None, image_size, image_size, num_channels])
- #DROPOUT
- #is_training = tf.placeholder(tf.bool)
- #CONV1
- W_conv1 = weight_variable([3, 3, 1, 128])
- b_conv1 = bias_variable([128])
- h_conv1 = tf.nn.relu(conv2d(xs, W_conv1) + b_conv1)
- #CONV2
- W_conv2 = weight_variable([3, 3, 128, 128])
- b_conv2 = bias_variable([128])
- h_conv2 = tf.nn.relu(conv2d(h_conv1, W_conv2) + b_conv2)
- #POOL1
- h_pool1 = max_pool_2x2(h_conv2)
- h_pool1_drop = h_pool1#tf.cond(is_training, lambda: tf.nn.dropout(h_pool1, keep_prob = 0.9), lambda: h_pool1)
- #CONV3
- W_conv3 = weight_variable([3, 3, 128, 128])
- b_conv3 = bias_variable([128])
- h_conv3 = tf.nn.relu(conv2d(h_pool1_drop, W_conv3) + b_conv3)
- #POOL2
- h_pool2 = max_pool_2x2(h_conv3)
- h_pool2_drop = h_pool2#tf.cond(is_training, lambda: tf.nn.dropout(h_pool2, keep_prob = 0.8), lambda: h_pool2)
- #LCONV1
- W_lconv1 = weight_variable([3, 3, 128, 256])
- b_lconv1 = bias_variable([256])
- h_lconv1 = tf.nn.relu(conv2d(h_pool2_drop, W_lconv1) + b_lconv1)
- #POOL3
- h_pool3 = max_pool_2x2(h_lconv1)
- h_pool3_drop = h_pool3#tf.cond(is_training, lambda: tf.nn.dropout(h_pool3, keep_prob = 0.7), lambda: h_pool3)
- #LCONV2
- W_lconv2 = weight_variable([3, 3, 256, 256])
- b_lconv2 = bias_variable([256])
- h_lconv2 = tf.nn.relu(conv2d(h_pool3_drop, W_lconv2) + b_lconv2)
- #POOL4
- h_pool4 = max_pool_2x2(h_lconv2)
- h_pool4_drop = h_pool4#tf.cond(is_training, lambda: tf.nn.dropout(h_pool4, keep_prob = 0.6), lambda: h_pool4)
- #LCONV3
- W_lconv3 = weight_variable([3, 3, 256, 256])
- b_lconv3 = bias_variable([256])
- h_lconv3 = tf.nn.relu(conv2d(h_pool4_drop, W_lconv3) + b_lconv3)
- #FLATTENING
- flat_h_pool4_drop = tf.reshape(h_pool4_drop, [-1, np.prod(h_pool4_drop.shape[1:]).value])
- flat_h_lconv3 = tf.reshape(h_lconv3, [-1, np.prod(h_lconv3.shape[1:]).value])
- flat_concat = tf.concat([flat_h_pool4_drop, flat_h_lconv3], 1)
- #FC1
- W_fc1 = weight_variable([flat_concat.get_shape().as_list()[1], descriptor_size])
- b_fc1 = bias_variable([descriptor_size])
- h_fc1 = tf.nn.bias_add(tf.matmul(flat_concat, W_fc1), b_fc1)# tf.matmul(flat_concat, W_fc1) + b_fc1
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement