我正在尝试对HiRISE数据集中的某些火星图像进行图像分类。运行代码时遇到以下错误
文件“ C：\ Users \ Anaconda3 \ lib \ site-packages \ tensorflow \ python \ ops \ gen_functional_ops.py”，行521，在stateful_partitioned_call中 _six.raise_from（_core._status_to_exception（e.code，message），None）
文件“”，第3行，位于raise_from
InvalidArgumentError：不兼容的形状：[32]与[32,56,56] [[{{nodemetrics_19 / acc / Equal}}]] [Op：StatefulPartitionedCall]
我为此代码安装了每个模块，但是似乎参数不匹配，尤其是关于“批量大小”
# Tensorflow imports
import tensorflow as tf
import tensorflow_datasets as tfds
tf.logging.set_verbosity(tf.logging.ERROR)
# Helper libraries
import math
import numpy as np
import matplotlib
matplotlib.use('PS') #prevent import error due to venv
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
# Imports for dataset separation
from sklearn.model_selection import train_test_split
# Improve progress bar display
import tqdm
import tqdm.auto
tqdm.tqdm = tqdm.auto.tqdm
# allow for dataset iteration.
tf.enable_eager_execution() #comment this out if causing errors
### GET THE DATASET AND SOME INFO ABOUT IT ###
# get the data into slices
data_images = []
data_labels = []
rel_img_path = 'map-proj/' # add path of folder to image name for later loadinr
# open up the labeled data file
with open('labels-map-proj.txt') as labels:
for line in labels:
file_name, label = line.split(' ')
data_images.append(rel_img_path + file_name)
data_labels.append(int(label))
# divide data into testing and training (total len 3820)
test_len = tf.cast(len(data_labels) * 0.15, tf.int64) # 573
train_len = tf.cast(len(data_labels) * 0.85, tf.int64) # 3247
train_images, test_images, train_labels, test_labels = train_test_split(
data_images, data_labels, test_size=0.15, random_state=666)
# convert training and testing data to tf dataset
train_dataset = tf.data.Dataset.from_tensor_slices((train_images, train_labels))
test_dataset = tf.data.Dataset.from_tensor_slices((test_images, test_labels))
# label translations
class_labels = ['other','crater','dark_dune','streak',
'bright_dune','impact','edge']
### PREPROCESS THE DATA ###
#convert image paths into real images
def parse_image(filename, label):
im_string = tf.read_file(filename)
im_decoded = tf.image.decode_jpeg(im_string, channels=1)
img = tf.cast(im_decoded, tf.float32)
return img, label
train_dataset = train_dataset.map(parse_image)
test_dataset = test_dataset.map(parse_image)
# convert to 0-1 range
def normalize(image, label):
image = tf.cast(image, tf.float32)
image /= 255.0
return image, label
train_dataset = train_dataset.map(normalize)
test_dataset = test_dataset.map(normalize)
### BUILD SHAPE OF THE MODEL ###
# increase kernel size and stride??
model = tf.keras.Sequential([
tf.keras.layers.Conv2D(32, (3,3), padding='same', activation=tf.nn.relu,
input_shape=(227,227,1)),
tf.keras.layers.MaxPooling2D((2,2), strides=2),
tf.keras.layers.Conv2D(64, (3,3), padding='same', activation=tf.nn.relu),
tf.keras.layers.MaxPooling2D((2,2), strides=2),
tf.keras.layers.Dense(128, activation=tf.nn.relu),
tf.keras.layers.Dense(7, activation=tf.nn.softmax) # final layer with node for each classification
])
# specify loss functions
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
### TRAIN THE MODEL ###
#specify training metadata
BATCH_SIZE = 32
# model.fit requires train data to be in the shape of [batch, imDim1, imDim2, numChannels]
#train_images = tf.reshape(train_images, [-1, 227, 227, 1])
#print("input shape is ", train_images.shape)
train_dataset = train_dataset.repeat().shuffle(train_len).batch(BATCH_SIZE)
test_dataset = test_dataset.batch(BATCH_SIZE)
print("about to train")
# train the model on the training data
num_epochs = 1 #TODO: increase later
model.fit(train_dataset, epochs=num_epochs, steps_per_epoch=math.ceil(train_len/BATCH_SIZE))
### EVALUATE MODEL ACCURACY ###
test_loss, test_accuracy = model.evaluate(test_dataset, steps=math.ceil(test_len/BATCH_SIZE))
print("Final loss was {}.\nAccuracy of model was {}".format(test_loss,test_accuracy))