Question

我有一个简单的Tensorflow模型，它是用python编写的，可以顺利训练。这是我正在使用的模型：

%template

我将模型导出到protobuf版本，然后将其加载到我的C ++代码中，以便使用C ++ API而不是Python来训练模型。不幸的是，我会收到有关SparseSoftmaxCrossEntropy操作的错误：

import tensorflow as tf from tensorflow.python.framework 
import ops from tensorflow.python.framework 
import dtypes

import random  import numpy as np

NUM_CLASSES = 102  IMAGE_HEIGHT = 224  IMAGE_WIDTH = 224  BATCH_SIZE = 25  NUM_CHANNELS = 3  LEARNING_RATE = 0.0001

with tf.Session() as sess:



images_placeholder = tf.placeholder (tf.float32,
                                         shape=(BATCH_SIZE, IMAGE_HEIGHT,
                                                IMAGE_WIDTH, NUM_CHANNELS), name="input")   labels_placeholder = tf.placeholder (tf.float32,
                                         shape=(BATCH_SIZE), name="label")

    with tf.name_scope("conv1_1") as scope:         kernel = tf.Variable (tf.truncated_normal([3, 3, 3, 64], dtype=tf.float32, stddev=1e-2),
                                name="weights")         conv = tf.nn.conv2d (images_placeholder, kernel, [1, 1, 1, 1], padding='SAME')      biases = tf.Variable (tf.constant(0.0, shape=[64], dtype=tf.float32),
                                trainable=True, name='biases')      out = tf.nn.bias_add (conv, biases)         conv1_1 = tf.nn.relu (out, name=scope)

    pool1 = tf.nn.max_pool (conv1_1,
                            ksize=[1, 2, 2, 1],
                            strides=[1, 2, 2, 1],
                            padding='SAME',
                            name='pool1')

    with tf.name_scope('conv2_1') as scope:         kernel = tf.Variable (tf.truncated_normal([3, 3, 64, 128], dtype=tf.float32, 
                                stddev=1e-2), name='weights')       conv = tf.nn.conv2d (pool1, kernel, [1, 1, 1, 1], padding='SAME')       biases = tf.Variable (tf.constant(0.0, shape=[128], dtype=tf.float32),
                                trainable=True, name='biases')      out = tf.nn.bias_add (conv, biases)         conv2_1 = tf.nn.relu (out, name=scope)

    pool2 = tf.nn.max_pool (conv2_1,
                            ksize=[1, 2, 2, 1],
                            strides=[1, 2, 2, 1],
                            padding='SAME',
                            name='pool2')

    with tf.name_scope('conv3_1') as scope:         kernel = tf.Variable(tf.truncated_normal([3, 3, 128, 256], dtype=tf.float32,
                                    stddev=1e-2), name='weights')       conv = tf.nn.conv2d(pool2, kernel, [1, 1, 1, 1], padding='SAME')        biases = tf.Variable(tf.constant(0.0, shape=[256], dtype=tf.float32),
                                    trainable=True, name='biases')      out = tf.nn.bias_add(conv, biases)      conv3_1 = tf.nn.relu(out, name=scope)

    pool3 = tf.nn.max_pool (conv3_1,
                            ksize=[1, 2, 2, 1],
                            strides=[1, 2, 2, 1],
                            padding='SAME',
                            name='pool3')

    with tf.name_scope('conv4_1') as scope:         kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 512], dtype=tf.float32,
                                stddev=1e-2), name='weights')       conv = tf.nn.conv2d(pool3, kernel, [1, 1, 1, 1], padding='SAME')        biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
                                trainable=True, name='biases')      out = tf.nn.bias_add(conv, biases)      conv4_1 = tf.nn.relu(out, name=scope)

    pool4 = tf.nn.max_pool (conv4_1,
                            ksize=[1, 2, 2, 1],
                            strides=[1, 2, 2, 1],
                            padding='SAME',
                            name='pool4')   

    with tf.name_scope('mentee_conv5_1') as scope:      kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512], dtype=tf.float32,
                                stddev=1e-2), name='weights')       conv = tf.nn.conv2d(pool4, kernel, [1, 1, 1, 1], padding='SAME')        biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
                                 trainable=True, name='biases')         out = tf.nn.bias_add(conv, biases)      conv5_1 = tf.nn.relu(out, name=scope)

    pool5 = tf.nn.max_pool (conv5_1,
                            ksize=[1, 2, 2, 1],
                            strides=[1, 2, 2, 1],
                            padding='SAME',
                            name='pool5')

    with tf.name_scope('fc1') as scope:         shape = int(np.prod(pool5.get_shape()[1:]))         fc1w = tf.Variable(tf.truncated_normal([shape, 4096], dtype=tf.float32, 
                                stddev=1e-2), name='weights')       fc1b = tf.Variable(tf.constant(1.0, shape=[4096], dtype=tf.float32),
                                trainable=True, name='biases')      pool5_flat = tf.reshape(pool5, [-1, shape])         fc1l = tf.nn.bias_add(tf.matmul(pool5_flat, fc1w), fc1b)        fc1 = tf.nn.relu(fc1l)
        fc1 = tf.nn.dropout(fc1, 0.5)


    labels = tf.to_int64(labels_placeholder)    cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits (labels=labels,
                        logits=fc1, name="xentropy")    loss = tf.reduce_mean (cross_entropy, name='loss')

    optimizer = tf.train.AdamOptimizer (LEARNING_RATE)  global_step = tf.Variable (0, name='global_step', trainable=False)  train_op = optimizer.minimize (loss, global_step=global_step, name="train")

    init = tf.initialize_variables (tf.all_variables(), name='init_all_vars_op')    tf.train.write_graph (sess.graph_def, "models/", "graph.pb", as_text=False)

我检查了tensorflow源代码，并意识到此操作的梯度计算尚未实现。但我想知道python版本如何成功没有问题，但在C ++中它试图计算渐变？

Tensorflow无法计算操作梯度

0 个答案: