python - How can I solve Value error in resnet 50 implementation?

Question

I am implementing resnet-50 on Kaggle and I am getting a value error. Kindly help me out

    train_dir='../input/project/data/train'
    test_dir='../input/project/data/test'
    
    train_datagen=ImageDataGenerator(rescale=1./255,
          rotation_range=40,
          width_shift_range=0.2,
          height_shift_range=0.2,
          shear_range=0.2,
          zoom_range=0.2,
          horizontal_flip=True,
          fill_mode='nearest')
    test_datagen = ImageDataGenerator(rescale = 1./255)
    
    train_generator = train_datagen.flow_from_directory(
        train_dir,
      color_mode='grayscale',
        target_size=(28,28),
        class_mode='binary',
      batch_size=32,
    )
    test_generator = test_datagen.flow_from_directory(
        test_dir,
      color_mode='grayscale',
        target_size=(28,28),
        class_mode='binary',
      batch_size=32,
        shuffle='False',
        
    )
    model = Sequential()
    
    model.add(ResNet50(include_top=False, pooling='avg', weights=resnet_weights_path,input_tensor=Input(shape=(224,224,3))))
    model.add(Flatten())
    model.add(BatchNormalization())
    model.add(Dense(2048, activation='relu'))
    model.add(BatchNormalization())
    model.add(Dense(1024, activation='relu'))
    model.add(BatchNormalization())
    model.add(Dense(2, activation='sigmoid'))
    
    model.layers[0].trainable = False

I am training a binary classifier and I am getting the error below

ValueError: Cannot assign to variable conv3_block1_0_conv/kernel:0 due to variable shape (1, 1, 256, 512) and value shape (512, 128, 1, 1) are incompatible

question from:https://stackoverflow.com/questions/66049551/how-can-i-solve-value-error-in-resnet-50-implementation

Answer 1

You have given input_tensor=Input(shape=(224,224,3)) while defining the ResNet50 base model. But you are giving target_size=(28,28) in your train_generator and test_generator. The training image shape which ResNet50 receiving i.e. target_size is different from what it expects i.e. input_tensor. Change your target_size to match with the shape mentioned in the input_tensor. Also, ResNet50 expects color_mode to be rgb rather grayscale.