Question

我有输入函数，它从filename参数中读取所有行。然后它将解析器函数应用于每一行。我的.csv文件中的每一行如下：＆＃34; path_to_img，label_num＆＃34;。这是代码：

def input_fn(filename):

    def parser(column):
        d = tf.decode_csv(column, record_defaults=[[''], [-1]])
        img, label = d
        img = tf.read_file(img)
        img = tf.image.decode_jpeg(img, channels=3)
        img = tf.image.resize_images(img, (300, 200))
        img = tf.cast(img, tf.float32)
        return {'image': img}, label

    dataset = tf.data.TextLineDataset(filename).map(parser)
    dataset = dataset.repeat(None).shuffle(5000).batch(500)
    return dataset.make_one_shot_iterator().get_next()

然后我创建分类器并训练：

feat_cols = [tf.feature_column.numeric_column('image', shape=(300, 200))]

model = tf.estimator.DNNClassifier([4028, 1024, 512], feature_columns=feat_cols, model_dir=OUTDIR, n_classes=5)

model.train(input_fn=lambda: input_fn('data.csv'), steps=1000)

但最终我得到的错误与解码操作有关，但仍无法理解错误

tensorflow.python.framework.errors_impl.InvalidArgumentError: Expected image (JPEG, PNG, or GIF), got unknown format starting with '\"\"\"\n    flickr.p'
 [[Node: DecodeJpeg = DecodeJpeg[acceptable_fraction=1, channels=3, dct_method="", fancy_upscaling=true, ratio=1, try_recover_truncated=false](ReadFile)]]
 [[Node: IteratorGetNext = IteratorGetNext[output_shapes=[[?,300,200,1], [?]], output_types=[DT_FLOAT, DT_INT32], _device="/job:localhost/replica:0/task:0/device:CPU:0"](OneShotIterator)]]

张量流中的图像预处理错误

0 个答案: