如何使用Keras与Python加载、展示猫狗数据集图像及解决路径查询空列表问题

Hey there! Let's fix that empty list issue first, then walk through how to load and display the cat/dog images properly.

1. Why cats = list(train_dir.glob('cats/*')) returns an empty list? #

The problem boils down to incorrect path hierarchy plus a tiny typo:

• The cats_and_dogs_filtered.zip you downloaded unzips into a root folder named cats_and_dogs_filtered, which contains the actual train and validation subfolders inside it.
• In your code, tf.keras.utils.get_file returns the path to the extracted outer folder (like /root/.keras/datasets/cats_and_dogs_filterted), but you're looking for train directly under this path—when in reality, train lives one level deeper inside the inner cats_and_dogs_filtered folder.
• Also, you misspelled the dataset name: the original is cats_and_dogs_filtered (ends with r), but you used cats_and_dogs_filterted (missing the final r) in get_file, which adds to the path confusion.

Here's the fixed code:

import tensorflow as tf
import pathlib
import os

# Fix the filename spelling to match the original dataset
_URL = 'https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip'
data_dir = tf.keras.utils.get_file('cats_and_dogs_filtered', origin=_URL, extract=True)
# Point to the actual dataset folder inside the extracted directory
data_dir = pathlib.Path(data_dir).parent / 'cats_and_dogs_filtered'

train_dir = data_dir / 'train'
validation_dir = data_dir / 'validation'

# Test if we can find cat images now
cats = list(train_dir.glob('cats/*'))
print(len(cats))  # Should output 1000—success!

Alternatively, you can dynamically find the inner folder using glob:

data_dir = pathlib.Path(data_dir)
data_dir = next(data_dir.glob('cats_and_dogs_filtered'))

2. Loading and displaying cat/dog images with Keras & Python #

Let's cover two common methods: loading a single image directly, and using Keras's ImageDataGenerator to batch-load and visualize the dataset.

Method 1: Load and display a single image

We'll use PIL and matplotlib for this—super straightforward:

import matplotlib.pyplot as plt
from PIL import Image

# Grab the first cat image path
cat_img_path = cats[0]
img = Image.open(cat_img_path)

# Show the image
plt.figure(figsize=(8, 8))
plt.imshow(img)
plt.title('A Cute Cat!')
plt.axis('off')  # Hide axes for cleaner look
plt.show()

Method 2: Batch-load and display images with ImageDataGenerator

ImageDataGenerator is Keras's go-to tool for handling image datasets—it handles resizing, scaling, and even data augmentation. Here's how to use it:

from tensorflow.keras.preprocessing.image import ImageDataGenerator

# Initialize the generator (we'll scale pixel values to 0-1 first)
train_datagen = ImageDataGenerator(rescale=1./255)

# Load training data from the directory
train_generator = train_datagen.flow_from_directory(
    train_dir,
    target_size=(150, 150),  # Resize all images to 150x150
    batch_size=32,
    class_mode='binary'  # Binary classification (cat vs dog)
)

# Get one batch of images and their labels
images, labels = next(train_generator)

# Display the first 9 images in the batch
plt.figure(figsize=(10, 10))
for i in range(9):
    plt.subplot(3, 3, i+1)
    plt.imshow(images[i])
    # Label based on the generator's class order (0 = cat, 1 = dog, alphabetical)
    plt.title('Cat' if labels[i] == 0 else 'Dog')
    plt.axis('off')
plt.show()

Quick note: flow_from_directory automatically assigns labels based on subfolder names, sorted alphabetically—so cats gets label 0, dogs gets label 1.

Bonus: Add data augmentation (for better model training) #

If you want to augment your data to improve model generalization, just tweak the ImageDataGenerator setup:

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,  # Rotate images up to 40 degrees
    width_shift_range=0.2,  # Shift horizontally by 20% of width
    height_shift_range=0.2,  # Shift vertically by 20% of height
    shear_range=0.2,  # Shear transformation
    zoom_range=0.2,  # Zoom in/out by 20%
    horizontal_flip=True,  # Flip images horizontally
    fill_mode='nearest'  # Fill in missing pixels after transformation
)

For larger datasets, you can also use tf.data.Dataset (available in TensorFlow 2.4+) for faster loading and more control.

内容的提问来源于stack exchange，提问作者Osama Mohammed