自定义超分辨率图像HDF5数据集保存后无法显示图像求助

我正在为超分辨率深度学习网络开发自定义图像数据集，已将图像保存到磁盘并能生成HDF5数据集文件，使用的代码如下：

import os, cv2, h5py, glob
import numpy as np
from glob import glob

# define the paths to the dataset
BASE_DATA_PATH = '/usr/local/home/.../esrgan_data'
HR_TRAIN_PATH = os.path.join(BASE_DATA_PATH, 'train_HR')
LR_TRAIN_PATH = os.path.join(BASE_DATA_PATH, 'train_LR')

# create LR and HR image lists
LR_images = glob(LR_TRAIN_PATH + '**/*.png')
HR_images = glob(HR_TRAIN_PATH + '**/*.png')

# sort the lists
LR_images.sort()
HR_images.sort()

# create an h5 file
with h5py.File('datasets/esrgan_trainDS.h5', 'w') as h5_file:
    # create 2 datasets for LR and HR images in the h5 file
    lr_ds = h5_file.create_dataset('trainLR', (len(LR_images), 150, 150, 3), dtype='f')
    hr_ds = h5_file.create_dataset('trainHR', (len(HR_images), 600, 600, 3), dtype='f')
    
    for i in range(len(LR_images)):
        LR_image = cv2.imread(LR_images[i])
        HR_image = cv2.imread(HR_images[i])
        lr_trainDS[i] = LR_image
        hr_trainDS[i] = HR_image

# load the h5 dataset
trainDS = h5py.File('datasets/esrgan_trainDS.h5', 'r')
print('Files in the training dataset: ', list(trainDS.keys()))

运行后输出：

Files in the training dataset:  ['trainHR', 'trainLR']

执行以下代码查看数据集形状：

LRset = trainDS['trainLR']
HRset = trainDS['trainHR']

print('LR dataset shape: ', LRset.shape)
print('HR dataset shape: ', HRset.shape)

输出结果：

LR dataset shape:  (450, 150, 150, 3)
HR dataset shape:  (450, 600, 600, 3)

但尝试查看数据集中的单张图像时，显示为全黑框：

cv2_imshow('', HRset[100])

代码无报错，可正常读写HDF5文件并打印属性，但无法正常显示图像。

问题排查与修复方案 #

1. 核心问题：变量名笔误导致数据未写入

代码中创建HDF5数据集时用的变量是lr_ds和hr_ds，但循环写入数据时错误使用了未定义的lr_trainDS和hr_trainDS。这会导致HDF5数据集仅被初始化了形状，但实际没有写入任何图像数据，所有值都是默认的0，因此显示为黑框。

2. 数据类型与显示格式问题

• cv2读取的图像是uint8类型（像素值范围0-255），直接存储为float类型会导致数据范围不匹配，建议归一化到0-1区间；
• 读取显示时需要将float类型转换回uint8类型，同时注意cv2默认读取的是BGR格式，显示时可转换为RGB适配常规显示逻辑。

修复后的完整代码

import os, cv2, h5py, glob
import numpy as np
from glob import glob

# define the paths to the dataset
BASE_DATA_PATH = '/usr/local/home/.../esrgan_data'
HR_TRAIN_PATH = os.path.join(BASE_DATA_PATH, 'train_HR')
LR_TRAIN_PATH = os.path.join(BASE_DATA_PATH, 'train_LR')

# create LR and HR image lists（添加recursive=True确保遍历子文件夹）
LR_images = glob(LR_TRAIN_PATH + '**/*.png', recursive=True)
HR_images = glob(HR_TRAIN_PATH + '**/*.png', recursive=True)

# sort the lists
LR_images.sort()
HR_images.sort()

# 校验LR和HR图像数量是否匹配
assert len(LR_images) == len(HR_images), "LR和HR图像数量不匹配"

# create an h5 file
with h5py.File('datasets/esrgan_trainDS.h5', 'w') as h5_file:
    # create 2 datasets for LR and HR images in the h5 file
    lr_ds = h5_file.create_dataset('trainLR', (len(LR_images), 150, 150, 3), dtype='float32')
    hr_ds = h5_file.create_dataset('trainHR', (len(HR_images), 600, 600, 3), dtype='float32')
    
    for i in range(len(LR_images)):
        # 读取图像，确保读取成功
        LR_image = cv2.imread(LR_images[i])
        HR_image = cv2.imread(HR_images[i])
        assert LR_image is not None, f"无法读取LR图像：{LR_images[i]}"
        assert HR_image is not None, f"无法读取HR图像：{HR_images[i]}"
        
        # 归一化到0-1范围，转换为float32
        lr_ds[i] = LR_image.astype(np.float32) / 255.0
        hr_ds[i] = HR_image.astype(np.float32) / 255.0

# load the h5 dataset
trainDS = h5py.File('datasets/esrgan_trainDS.h5', 'r')
print('Files in the training dataset: ', list(trainDS.keys()))

LRset = trainDS['trainLR']
HRset = trainDS['trainHR']

print('LR dataset shape: ', LRset.shape)
print('HR dataset shape: ', HRset.shape)

# 查看图像的正确方式
sample_hr = HRset[100]
# 将0-1的float转换回0-255的uint8，并转换BGR到RGB
sample_hr = (sample_hr * 255).astype(np.uint8)
sample_hr_rgb = cv2.cvtColor(sample_hr, cv2.COLOR_BGR2RGB)
cv2_imshow('Sample HR Image', sample_hr_rgb)

内容的提问来源于stack exchange，提问作者west42