keras 输入层限制

Question

example model from a book goes well.

model = keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28, 28]))
model.add(keras.layers.Dense(300, activation="relu"))
model.add(keras.layers.Dense(100, activation="relu"))
model.add(keras.layers.Dense(10, activation="softmax"))

Though when I try to change input_shape from [28, 28] to [1920, 1080] or [1,] compiler says this:

File "C:\Users\User1\PycharmProjects\untitled3\venv\lib\site-packages\tensorflow\python\framework\ops.py", line 6862, in raise_from_not_ok_status six.raise_from(core._status_to_exception(e.code, message), None) File "", line 3, in raise_from

tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor with shape[2073600,300] and type float on /job:localhost/replica:0/task:0/device:CPU:0 by allocator cpu [Op:Mul]

for me it looks like it loads tensor in processor, but it is not data, just model creation. what problem can it be?

Answer 1

This is why convolutional neural networks should be used when dealing with images. Every pixel becomes a feature, and every pixel will be connected to every unit of your dense layer.

Let's say you have a 1920*1080 pictures and a dense layer with 300 units, that's 622,080,000 parameters just for your first layer, which will overload your GPU. Try using a convnet instead.

model = keras.Sequential(
    [
        keras.Input(shape=input_shape),
        layers.Conv2D(32, kernel_size=(3, 3), activation="relu"),
        layers.MaxPooling2D(pool_size=(2, 2)),
        layers.Conv2D(64, kernel_size=(3, 3), activation="relu"),
        layers.MaxPooling2D(pool_size=(2, 2)),
        layers.Flatten(),
        layers.Dropout(0.5),
        layers.Dense(num_classes, activation="softmax"),
    ]
)