使用Keras训练CNN手语识别模型出现形状不兼容ValueError报错

云原生阿岚

2026-8-29

问题背景

开发摄像头驱动的手语翻译工具时，基于Keras搭建CNN模型训练手部手势识别数据集，训练过程触发如下报错：

ValueError: (500,44) and (500,40) are incompatible

报错根因

该错误为张量形状不匹配导致，第一维500是设置的batch size，维度匹配无问题；第二维是类别维度，不匹配的核心原因有两个：

代码混用了独立Keras包和TensorFlow内置的tf.keras接口，两套接口存在版本兼容差异，会导致层输出形状计算、字段名逻辑不一致。
尺寸、类别数的获取逻辑不可靠：原代码通过读取单张样本图获取输入尺寸、通过遍历手势文件夹数量获取类别数，但序列化存储在pickle文件中的训练/验证集是之前生成的，如果后续调整过图片尺寸、增减过手势类别文件夹但没有重新生成pickle数据集，就会出现模型输出维度（44维，对应遍历得到的44个类别文件夹）和标签one-hot编码后的实际维度（40维，对应数据集生成时的40个类别）不匹配，计算交叉熵损失时直接抛错。
附带隐性问题：代码中ModelCheckpoint监控的val_acc、SGD优化器的lr参数都是旧版Keras的命名，新版TensorFlow中已经变更，就算形状问题修复也会触发警告或异常。

可落地修复方案

按以下步骤调整代码即可解决问题：

统一Keras导入来源，全部使用TensorFlow内置的keras接口，替换原有导入段：

import numpy as np
import pickle
import cv2, os
from glob import glob
from tensorflow.keras import optimizers
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras import backend as K
from tensorflow.keras.utils import plot_model
K.set_image_data_format('channels_last')
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

调整模型构建函数，改为接收外部传入的固定图片尺寸、类别数参数，同时修正旧版参数命名问题：

def cnn_model(image_x, image_y, num_of_classes):
    model = Sequential()
    model.add(Conv2D(16, (2,2), input_shape=(image_x, image_y, 1), activation='relu'))
    model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'))
    model.add(Conv2D(32, (3,3), activation='relu'))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(3, 3), padding='same'))
    model.add(Conv2D(64, (5,5), activation='relu'))
    model.add(MaxPooling2D(pool_size=(5, 5), strides=(5, 5), padding='same'))
    model.add(Flatten())
    model.add(Dense(128, activation='relu'))
    model.add(Dropout(0.2))
    model.add(Dense(num_of_classes, activation='softmax'))
    # 新版SGD学习率参数名为learning_rate，替换旧版lr
    sgd = optimizers.SGD(learning_rate=1e-2)
    model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
    filepath="cnn_model_keras2.h5"
    # 新版验证集精度字段名为val_accuracy，替换旧版val_acc
    checkpoint1 = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')
    callbacks_list = [checkpoint1]
    plot_model(model, to_file='model.png', show_shapes=True)
    return model, callbacks_list

调整训练逻辑，直接从加载的pickle数据集中获取真实图片尺寸、类别数，不再依赖单图读取、文件夹遍历的动态获取逻辑，避免数据不一致：

def train():
    with open("train_images", "rb") as f:
        train_images = np.array(pickle.load(f))
    with open("train_labels", "rb") as f:
        train_labels = np.array(pickle.load(f), dtype=np.int32)

    with open("val_images", "rb") as f:
        val_images = np.array(pickle.load(f))
    with open("val_labels", "rb") as f:
        val_labels = np.array(pickle.load(f), dtype=np.int32)

    # 从加载的训练集直接取真实图片尺寸
    img_size = train_images.shape[1]
    train_images = np.reshape(train_images, (train_images.shape[0], img_size, img_size, 1))
    val_images = np.reshape(val_images, (val_images.shape[0], img_size, img_size, 1))
    
    # 从标签数组直接统计真实类别数
    num_classes = len(np.unique(train_labels))
    train_labels = to_categorical(train_labels, num_classes=num_classes)
    val_labels = to_categorical(val_labels, num_classes=num_classes)

    print(val_labels.shape)
    model, callbacks_list = cnn_model(img_size, img_size, num_classes)
    model.summary()
    model.fit(train_images, train_labels, validation_data=(val_images, val_labels), epochs=20, batch_size=500, callbacks=callbacks_list)
    scores = model.evaluate(val_images, val_labels, verbose=0)
    print("CNN Error: %.2f%%" % (100-scores[1]*100))
    model.save('cnn_model_keras2.h5')

删除原有全局定义的image_x, image_y = get_image_size()、get_image_size()、get_num_of_classes()冗余代码，执行脚本即可正常训练。

内容的提问来源于stack exchange，提问作者ArkamC