KMeans图像聚类循环失效：标签与图片数量不匹配问题求助

问题场景 #

使用GitHub仓库study2A的代码进行图像聚类时，KMeans聚类环节无法为每张图片分配标签——生成的标签数量为9，但图片数量为200，二者长度不匹配，无法完成标签赋值并生成带标签的txt文件。

原代码 #

import joblib, os
import numpy as np
import pandas as pd
from sklearn.cluster import KMeans
import ypoften as of

cwd = os.path.join('/content/drive/MyDrive/study2A/for lab teaching',"")
# please change this to the folder that has all the images

cvmodels = ["vgg16 fc1"]
## cvmodels = ["vgg16 fc1","vgg16 places fc1","vggface fc6"] choose the model you used
clmethod = "KMeans"

for cvmodel in cvmodels:

    features_savepath = os.path.join(cwd,'img exfeature1','features PCA',cvmodel+'.dat')
    features_array = joblib.load(features_savepath)
    X = pd.DataFrame(features_array)
    print(X)

    nd = 200
    X = X.iloc[:,0:nd]

    savefolder = cvmodel + ' ' + clmethod + ' PCA' + str(nd)

    # Read the 0321_imageselected_A.txt file
    imgnamefile = os.path.join(cwd, "0321_imageselected_A.txt")
    df = pd.read_csv(imgnamefile, sep='\t', header=0)

    for K in [2]:
        print('number of cluster', K)
        cl = KMeans(K, random_state=0)
        cl.fit(X)
        labels = cl.labels_

        imgnamefile = os.path.join(cwd,"0321_imageselected_A.txt")
        df = pd.read_csv(imgnamefile, sep ='\t', header = 0)
        print(df)

        df['label'] = labels
        filepath = os.path.join(cwd,'img cluster',savefolder,str(K),'label.txt')
        of.create_path(filepath)
        df.to_csv(filepath, index = None, header = None, sep = '\t')

    print("DONE"*20)

报错信息 #

ValueError: Length of values (9) does not match length of index (200)

原因分析 #

报错的核心是加载的特征矩阵features_array仅包含9个样本的特征，而图片列表0321_imageselected_A.txt中有200张图片。KMeans聚类后生成的labels长度等于特征矩阵的行数（9），但图片数据框df的行数是200，赋值时必然出现长度不匹配。

修复步骤 #

1. 验证特征矩阵样本数：在加载特征文件后，打印特征矩阵的形状，确认样本数是否与图片数量一致。
2. 排查特征提取环节：如果特征矩阵样本数确实为9，说明特征提取阶段未正确为200张图片生成特征，需要回到特征提取代码中检查逻辑，确保生成的.dat文件包含200个样本的特征数据。
3. 优化代码冗余：原代码中重复读取图片列表文件，可简化以提升效率。

修复后代码 #

import joblib, os
import numpy as np
import pandas as pd
from sklearn.cluster import KMeans
import ypoften as of

cwd = os.path.join('/content/drive/MyDrive/study2A/for lab teaching',"")

cvmodels = ["vgg16 fc1"]
clmethod = "KMeans"

for cvmodel in cvmodels:
    features_savepath = os.path.join(cwd,'img exfeature1','features PCA',cvmodel+'.dat')
    features_array = joblib.load(features_savepath)
    
    # 关键：打印特征矩阵形状，验证样本数是否为200
    print(f"特征矩阵形状（样本数，特征数）：{features_array.shape}")
    if features_array.shape[0] != 200:
        print("警告：特征样本数与图片数量不匹配，请检查特征提取环节！")
        continue

    X = pd.DataFrame(features_array)
    nd = 200
    X = X.iloc[:,0:nd]

    savefolder = cvmodel + ' ' + clmethod + ' PCA' + str(nd)

    # 仅读取一次图片列表
    imgnamefile = os.path.join(cwd, "0321_imageselected_A.txt")
    df = pd.read_csv(imgnamefile, sep='\t', header=0)

    for K in [2]:
        print('聚类数量:', K)
        cl = KMeans(K, random_state=0)
        cl.fit(X)
        labels = cl.labels_

        # 确认标签长度与图片数量一致
        print(f"标签长度: {len(labels)}, 图片数量: {len(df)}")
        df['label'] = labels

        filepath = os.path.join(cwd,'img cluster',savefolder,str(K),'label.txt')
        of.create_path(filepath)
        df.to_csv(filepath, index=None, header=None, sep='\t')

    print("DONE"*20)

注意事项 #

• 如果打印特征矩阵形状后发现样本数不是200，必须优先修正特征提取流程，确保每个图片对应一行特征数据。
• 若特征矩阵样本数正确但仍报错，可检查图片列表文件是否存在重复行或读取错误。

内容的提问来源于stack exchange，提问作者user23713357