使用卷积神经网络进行多标签图像分类

Question

我尝试为图像数据集创建一个CNN网络，该数据集包含20个文件夹中总共2280张图像（每个文件夹包含114张图像）。我已经处理了图像，从文件夹中读取图像并创建了字典。但是，当我想提供整个数据集以将数据拆分为训练和测试时，我得到的标签大小不正确。图像数组的大小是正确的：(2280,56,56)，但 all_labels 数组应该是 (2280,20)，而我得到 (2280,56)

如果有人可以帮助我，那将非常感激。 提前谢谢您！

所以我的代码如下

def read_image(image_path):
    # Read the image using OpenCV
    image = cv2.imread(image_path)
    # Crop the image
    #image_crop = image[:, left_crop:image.shape[1] - right_crop]
    #cv2.imshow("output", image_crop)
    #cv2.waitKey()

    # Resize the image
    #image_resized = cv2.resize(image_crop, (width, height), interpolation=cv2.INTER_AREA)
        
    image_resized = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        
    # Normalize the pixel values
    image_resized = image_resized.astype('float32') / 255.0
    #print(len(image_resized.shape))

    # Print the size of the image resized
    #print(f"Image resize size: {image_resized.shape}")

    #cv2.namedWindow("output", cv2.WINDOW_NORMAL) 
    #cv2.imshow("output", image_resized)
    #cv2.waitKey()

    return image_resized

def process_folder(folder_path):
    images = []

    # Get a list of all files (images) in the folder
    image_files = sorted([f for f in os.listdir(folder_path) if os.path.isfile(os.path.join(folder_path, f))])

    for image_file in image_files:
        image_path = os.path.join(folder_path, image_file)

        # Crop and resize the image
        image_processed = read_image(image_path)

        # Append to the list
        images.append(image_processed)


    # Convert the list of images to a numpy array
    images_array = np.array(images)

    return images_array

def create_folder_dictionary(root_folder_path):
    folder_dictionary = {}
    labels_data = {}

    # Get a list of all folders in the root folder
    folders = sorted([f for f in os.listdir(root_folder_path) if os.path.isdir(os.path.join(root_folder_path, f))])

    for folder in folders:
        
        folder_path = os.path.join(root_folder_path, folder)

        # Process each folder and store the resulting array in the dictionary
        folder_data = process_folder(folder_path)
        #folder_dictionary[folder] = folder_data
        if folder_data is not None:
            folder_dictionary[folder] = folder_data
            labels_data[folder] = folder_data

    return folder_dictionary,labels_data


root_folder_path = r'C:\Users\sumit\Downloads\master thesis\ImageDataset'

result,labels_data = create_folder_dictionary(root_folder_path)


# Extract keys and values
folders = list(result.keys())
folder_data_list = list(result.values())
#print(folders)
#print(len(folders))
#print(len(folder_data_list))

# Print the total number of images
total_images = sum(len(images) for images in folder_data_list)
print(f"Total number of images: {total_images}")


folders_new = list(map(int, folders))

folders_arr = np.array(folders_new)

#print("labels", folders_arr.shape)

folders_data_list_arr = np.array(folder_data_list)

#print("images", folders_data_list_arr.shape[2])

all_images = np.concatenate(folder_data_list, axis=0)
print(all_images.shape)



# Create all_labels 

all_labels = np.concatenate([np.full_like(data[:, 0], label) for label, data in zip(labels_data, folder_data_list)], axis=0)
 
 print("Shape of all_labels:", all_labels.shape)

Answer 1

在这一行中：

all_labels = np.concatenate([np.full_like(data[:, 0], label) for label, data in zip(labels_data, folder_data_list)], axis=0)

将 np.full_like(data[:, 0] 替换为 np.full(data.shape[0] 如果仍然出现错误，请尝试用 int(label) 替换该行中的 label 所以将该行替换为

all_labels = np.concatenate([np.full(data.shape[0], int(label)) for label, data in zip(labels_data, folder_data_list)], axis=0)

应该可以。