我一直在尝试训练模型。我写了一个脚本来构造图像数据集中的TFrecords。
这是训练脚本的代码：
import numpy as np
import tensorflow as tf
import cv2
from PIL import Image
import glob
import sys
batch_size=32
img_size=50
classes_names=[]
dataset_file=glob.glob("dataset/asl_alphabet_train/*")
sess = tf.Session()
sess.run(tf.global_variables_initializer())
for f in dataset_file:
classes_names.append(f[27:]) # the name of the subfolders is the name of the class
num_classes=len(classes_names)
def parser(record):
# a parsing function to parse the tfrecords
keys_to_features={
"img_raw": tf.FixedLenFeature([], tf.string),
"label": tf.FixedLenFeature([], tf.int64)
}
parsed=tf.parse_single_example(record, keys_to_features)#parsing one example from the example buffer from the tfrecord using the keys
image= tf.decode_raw(parsed["img_raw"], tf.uint8)# decoding ( bytes -> tf.float32)
image= tf.cast(image, tf.float32)
#image= tf.reshape(image, shape=[50,50,3])#reshaping images
label=tf.cast(parsed["label"], tf.int32)# casting labels to int32
return image, label
def input_fn(filenames, train_bool=True):
#from tfrecord to iterable data
dataset=tf.data.TFRecordDataset(filenames=filenames, num_parallel_reads=40)# instantiantion of an object from class TFRecordDataset
dataset= dataset.map(parser) # maps a function to the dataset
if train_bool:
dataset= dataset.shuffle(buffer_size=2048)
repeat=None # if in training mode allow reading data infinitely
else :
repeat=1 # if in validation or test allow max 1 read
dataset= dataset.repeat(repeat)
dataset= dataset.batch(batch_size)# define bach size
iterator= dataset.make_one_shot_iterator()# making the iterator
images_batch, labels_batch=iterator.get_next()# getting the data
x= {'image': images_batch}
y= labels_batch
return x, y
def train_input_fn():
return input_fn(filenames=["TFrecords/train.tfrecords"])
def val_input_fn():
return input_fn(filenames=["TFrecords/val.tfrecords"],train_bool=False)
def test_input_fn():
return input_fn(filenames=["TFrecords/test.tfrecords"])
def model_fn(features, labels, mode, params):
Input=features['image']# transform the features to a tensor
Input=tf.reshape(Input, [-1,img_size,img_size,3])# reshaping to tensorflow networks compatible shapes
output=tf.layers.conv2d(inputs=Input, name='conv1',
filters=32, kernel_size=3,
padding='same', activation=tf.nn.relu)
output=tf.layers.max_pooling2d(inputs=output, pool_size=2, strides=2)
output=tf.layers.conv2d(inputs=output, name='conv2',
filters=64, kernel_size=3,
padding='same', activation=tf.nn.relu)
output=tf.layers.max_pooling2d(inputs=output, pool_size=2, strides=2)
output=tf.layers.conv2d(inputs=output, name='conv3',
filters=64, kernel_size=3,
padding='same', activation=tf.nn.relu)
output=tf.layers.max_pooling2d(inputs=output, pool_size=2, strides=2)
output=tf.layers.flatten(output)
output=tf.layers.dense(inputs=output, name='fc1',
units=128, activation=tf.nn.relu)
output=tf.layers.dropout(inputs=output, rate=0.5, noise_shape=None,
seed=None, training=(mode==tf.estimator.ModeKeys.TRAIN))
output=tf.layers.dense(inputs=Input, name='fc2',units=num_classes)
logits=output
y_pred=tf.identity(tf.nn.softmax(logits=logits), name="logits")
y_pred_cls=tf.identity(tf.argmax(y_pred,axis=1), name="classes")
if mode == tf.estimator.ModeKeys.PREDICT:
spec = tf.estimator.EstimatorSpec(mode=mode,predictions=y_pred_cls)
else:
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits,labels=labels)
loss=tf.reduce_mean(cross_entropy)
optimizer= tf.train.AdamOptimizer(learning_rate=params["learning_rate"])
train_op= optimizer.minimize(loss=loss, global_step=tf.train.get_global_step())
metrics={ "accuracy":tf.metrics.accuracy(labels, y_pred_cls) }
spec=tf.estimator.EstimatorSpec(
mode=mode,
loss=loss,
train_op=train_op,
eval_metric_ops=metrics)
return spec
model = tf.estimator.Estimator(model_fn=model_fn,
params={"learning_rate":1e-2},
model_dir="Model/")
count=0
while(count<100000):
model.train(input_fn=train_input_fn,steps=1000)
result=model.evaluate(input_fn=val_input_fn)
print(count)
print("Classification accuracy: {0:.2%}".format(result["accuracy"]))
sys.stdout.flush()
count=count+1
运行脚本时，出现以下错误： ValueError：等级不匹配：标签的等级（接收到1）应等于对数的等级减去1（接收到4）。
此外，我尝试使用迭代器获取每批图像，并测试该批的大小，并测试是否收到正确的批大小，然后获得正确的批大小。但是，在模型的第一行中，当我尝试将张量整形为[batch_size，img_size，imge，size，3]时，我收到一个错误，即所接收的张量的数量是预期批处理数：0的4倍。
错误跟踪如下：
Traceback (most recent call last):
File "Train.py", line 122, in <module>
model.train(input_fn=train_input_fn,steps=1000)
File "/home/hch/miniconda3/envs/DL_CUDA/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 366, in train
loss = self._train_model(input_fn, hooks, saving_listeners)
File "/home/hch/miniconda3/envs/DL_CUDA/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 1119, in _train_model
return self._train_model_default(input_fn, hooks, saving_listeners)
File "/home/hch/miniconda3/envs/DL_CUDA/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 1132, in _train_model_default
features, labels, model_fn_lib.ModeKeys.TRAIN, self.config)
File "/home/hch/miniconda3/envs/DL_CUDA/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 1107, in _call_model_fn
model_fn_results = self._model_fn(features=features, **kwargs)
File "Train.py", line 101, in model_fn
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=labels,logits=logits)
File "/home/hch/miniconda3/envs/DL_CUDA/lib/python3.5/site-packages/tensorflow/python/ops/nn_ops.py", line 2052, in sparse_softmax_cross_entropy_with_logits
(labels_static_shape.ndims, logits.get_shape().ndims))
ValueError: Rank mismatch: Rank of labels (received 1) should equal rank of logits minus 1 (received 4).
您怎么看？ 谢谢
答案 0 :(得分：0)
我知道了！这是一个愚蠢的错误。在最后一个输出层，当我应该使用“输出”时，我输入“输入”：output = tf.layers.dense（inputs = Input，name ='fc2'，units = num_classes