ValueError: non-broadcastable output operand with shape (256,256,1) doesn't match the broadcast shape (256,256,3)

user_16175495 0
Viewed 202

I am using CNN to train a dataset of grayscale images (5 classes). The images are of size (100,100) and the pixel value is between 0-1 code :

from tensorflow.keras.preprocessing.image import ImageDataGenerator
import tensorflow as tf
import cv2

        
train = ImageDataGenerator(rescale=(100,100,1))
validation = ImageDataGenerator(rescale=(100,100,1))

train_dataset = train.flow_from_directory('C:/Users/abdul/OneDrive/Desktop/New folder/FYP/final/images/train',color_mode='grayscale')
validate_dataset = validation.flow_from_directory('C:/Users/abdul/OneDrive/Desktop/New folder/FYP/final/images/validate',color_mode='grayscale') 

model = tf.keras.models.Sequential([
    #convolutional layer
    tf.keras.layers.Conv2D(32,(3,3), activation ="relu", input_shape = (100,100,1)),
    
    # Flatten units
    tf.keras.layers.Flatten(),

    # Add a hidden layer with dropout
    tf.keras.layers.Dense(128, activation="relu"),
    tf.keras.layers.Dropout(0.5),

    # Add an output layer with output units for all 10 digits
    tf.keras.layers.Dense(4, activation="softmax")
])

# Train neural network
model.compile(
    optimizer="adam",
    loss="categorical_crossentropy",
    metrics=["accuracy"]
)

model.fit(train_dataset, epochs=10,validation_data = validate_dataset)
conv-neural-network keras tensorflow python
1 Answers

There is some issue with your data generator. In rescale you set wrongly the input size (probably) where you should set some factor that will be used to normalize your input. According to the doc:

rescale: rescaling factor. Defaults to None. If None or 0, no rescaling is applied, otherwise we multiply the data by the value provided (after applying all other transformations).

And, by default, the target_size is 256 x 256, but your model input is 100 x 100, so you need to specify that otherwise properly otherwise the data loader will generate an image with the shape of by its default value which is 256. According to the doc

target_size: tuple of integers (height, width), default: (256, 256). The dimensions to which all images found will be resized.

With these change, you should do as follows:

train = ImageDataGenerator(rescale=1/255.)
train_dataset = train.flow_from_directory('./train',
   color_mode='grayscale',
   class_mode='categorical', # one hot or 'sparse' if labels are integer
   target_size=(100, 100))

Do the same to the validation part.

user_92157800
Related