python |自动数据帧生成

Question

我有两个文件夹，其中包含两个不同白天(白天和夜晚)的城市天际线图像。我想读取相应文件夹中不同颜色空间的所有图像，然后我想计算所有颜色 channel 的统计数据。然后我想创建一个包含所有统计数据的 pandas 数据框。

为了防止不必要的重复代码，我尝试使用字典。目前，我可以打印颜色空间 x channel x 统计数据的所有组合的所有统计数据。但从概念上讲，我无法将这些东西放入包含行(单独的图像)和列(文件名、颜色空间 x channel x 统计数据)的 pandas DataFrame 中。

如果有任何帮助，我将不胜感激。

import os

import numpy as np
import matplotlib.pyplot as plt
import cv2
import pandas as pd


dictionary_of_color_spaces = {
    'RGB': cv2.COLOR_BGR2RGB,  # Red, Green, Blue
    'HSV': cv2.COLOR_BGR2HSV,  # Hue, Saturation, Value
    'HLS': cv2.COLOR_BGR2HLS,  # Hue, Lightness, Saturation
    'YUV': cv2.COLOR_BGR2YUV,  # Y = Luminance , U, V = Chrominance color components   
}

dictionary_of_channels = {
    'channel_1': 0,
    'channel_2': 1,
    'channel_3': 2,
}

dictionary_of_statistics = {
    'min': np.min,
    'max': np.max,
    'mean': np.mean,
    'median': np.median,
    'std': np.std,
}

# get filenames inside training folders for day and night
path_training_day = './day_night_images/training/day/'
path_training_night = './day_night_images/training/night/'
filenames_training_day = [file for file in os.listdir(path_training_day)]
filenames_training_night = [file for file in os.listdir(path_training_night)]

for filename in filenames_training_day:
    image = cv2.imread(path_training_day + filename)
    for color_space in dictionary_of_color_spaces:
        image = cv2.cvtColor(image, dictionary_of_color_spaces[color_space])
        for channel in dictionary_of_channels:
            for statistic in dictionary_of_statistics:
                print(dictionary_of_statistics[statistic](image[:,:,dictionary_of_channels[channel]]))

Answer 1

在不更改大部分代码的情况下，我能想到的最简单的事情是:

• 创建一个空的 df，其列都是统计数据 x channel x color_space 的组合(通过列表理解即可轻松完成)；
• 对于每个图像，将所有统计信息附加到一个变量(行):
• 将 row 转换为 pd.Series 对象，使用 row 作为值，使用数据帧的列作为索引，使用 filename 作为名称;
• 将该行附加到您的空 df 中。

最重要的细节是确保 df 列名称正确，即与填充 row 变量的值的顺序相同。当我们在列表理解中为列名创建组合时，重要的是我们从最内层循环移动到最外层，以便稍后在将 row 追加到 df 中时匹配值。

这应该有效:

import os

import numpy as np
import matplotlib.pyplot as plt
import cv2
import pandas as pd


dictionary_of_color_spaces = {
    'RGB': cv2.COLOR_BGR2RGB,  # Red, Green, Blue
    'HSV': cv2.COLOR_BGR2HSV,  # Hue, Saturation, Value
    'HLS': cv2.COLOR_BGR2HLS,  # Hue, Lightness, Saturation
    'YUV': cv2.COLOR_BGR2YUV,  # Y = Luminance , U, V = Chrominance color components   
}

dictionary_of_channels = {
    'channel_1': 0,
    'channel_2': 1,
    'channel_3': 2,
}

dictionary_of_statistics = {
    'min': np.min,
    'max': np.max,
    'mean': np.mean,
    'median': np.median,
    'std': np.std,
}

# creates column names in the same order as loops below
cols = [f'{s}_{c}_{cs}' for s in dictionary_of_statistics for c in dictionary_of_channels for cs in dictionary_of_color_spaces]
# creates empty df
df = pd.DataFrame(column=cols)


# get filenames inside training folders for day and night
path_training_day = './day_night_images/training/day/'
path_training_night = './day_night_images/training/night/'
filenames_training_day = [file for file in os.listdir(path_training_day)]
filenames_training_night = [file for file in os.listdir(path_training_night)]

for filename in filenames_training_day:
    row = []  # row for the current image - to be populated with stat values
    image = cv2.imread(path_training_day + filename)
    for color_space in dictionary_of_color_spaces:
        image = cv2.cvtColor(image, dictionary_of_color_spaces[color_space])
        for channel in dictionary_of_channels:
            for statistic in dictionary_of_statistics:
                row.append(dictionary_of_statistics[statistic](image[:,:,dictionary_of_channels[channel]]))
    row_series = pd.Series(row, index=cols, name=filename)
    df = df.append(row_series)

此代码将每个图像的文件名转换为最终 df 中每行的索引。如果您不想这样做，请将索引转换为新列 (df['filename'] = df.index) 并随后使用 pandas.reset_index (pd = pd.reset_index( drop=True).