Google Colab中TensorFlow代码目录找不到报错及代码问题求助

核心问题分析与修复步骤 #

1. 路径错误：Zip文件无法直接作为数据集目录 #

tf.keras.preprocessing.image_dataset_from_directory要求传入解压后的文件夹路径，而非压缩包路径。同时路径中的空格无需转义，需去掉多余的\符号。先解压Zip文件再指向文件夹：

# 解压Zip文件到指定目录
import zipfile
zip_ref = zipfile.ZipFile("/content/gdrive/MyDrive/Colab Notebooks/Pointing_dogs.zip", 'r')
zip_ref.extractall("/content/gdrive/MyDrive/Colab Notebooks/Pointing_dogs")
zip_ref.close()

# 指向解压后的文件夹
proj_path = "/content/gdrive/MyDrive/Colab Notebooks/Pointing_dogs"
proj_path = pathlib.Path(proj_path)

2. 数据集函数名与参数错误 #

验证集的函数名写错（少了directory后缀），且参数里的"proj_path"是字符串常量，需改为变量proj_path：

# 修正后的验证集代码
ds_validation = tf.keras.preprocessing.image_dataset_from_directory(
    proj_path,
    labels="inferred",
    label_mode="int",
    color_mode="grayscale",  # 和模型输入通道匹配
    batch_size=batch_size,
    image_size=(img_height, img_width),
    shuffle=True,
    seed=123,
    validation_split=0.2,
    subset="validation",
)

3. 输入通道与图像模式不匹配 #

模型输入定义为单通道灰度图(28, 28, 1)，但代码中color_mode设为"rgb"（3通道），会导致维度不匹配错误。需将训练集和验证集的color_mode统一改为"grayscale"，或把模型输入改为(28,28,3)（根据图像实际类型选择）。

4. 移除无用的自定义循环 #

代码中存在一段空的训练循环，会占用资源且与后续model.fit功能重复，直接删除即可：

# 删除这段无用代码
# for epochs in range(10):
#     for x, y in ds_train:
#         # train here
#         pass

完整修正后的代码 #

import os
import zipfile

os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"
import tensorflow as tf
import pathlib
from google.colab import drive
drive.mount('/content/gdrive')
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.preprocessing.image import ImageDataGenerator

img_height = 28
img_width = 28
batch_size = 2

# 解压Zip文件（仅当目录不存在时执行）
zip_path = "/content/gdrive/MyDrive/Colab Notebooks/Pointing_dogs.zip"
extract_path = "/content/gdrive/MyDrive/Colab Notebooks/Pointing_dogs"
if not os.path.exists(extract_path):
    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
        zip_ref.extractall(extract_path)

# 指向解压后的文件夹
proj_path = pathlib.Path(extract_path)

# 模型输入为单通道（匹配grayscale模式），若为RGB则改为(28,28,3)
model = keras.Sequential(
    [
        layers.Input((28, 28, 1)),
        layers.Conv2D(16, 3, padding="same"),
        layers.Conv2D(32, 3, padding="same"),
        layers.MaxPooling2D(),
        layers.Flatten(),
        layers.Dense(10),
    ]
)

# 训练集
ds_train = tf.keras.preprocessing.image_dataset_from_directory(
    proj_path,
    labels="inferred",
    label_mode="int",
    color_mode="grayscale",
    batch_size=batch_size,
    image_size=(img_height, img_width),
    shuffle=True,
    seed=123,
    validation_split=0.2,
    subset="training",
)

# 验证集
ds_validation = tf.keras.preprocessing.image_dataset_from_directory(
    proj_path,
    labels="inferred",
    label_mode="int",
    color_mode="grayscale",
    batch_size=batch_size,
    image_size=(img_height, img_width),
    shuffle=True,
    seed=123,
    validation_split=0.2,
    subset="validation",
)

# 数据增强
def augment(x, y):
    image = tf.image.random_brightness(x, max_delta=0.05)
    return image, y

ds_train = ds_train.map(augment)

# 编译与训练（加入验证集监控）
model.compile(
    optimizer=keras.optimizers.Adam(),
    loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    metrics=["accuracy"],
)

model.fit(ds_train, validation_data=ds_validation, epochs=2, verbose=2)

内容的提问来源于stack exchange，提问作者Jazaic Divina