Argthtjtr

I am trying to pass my own data from scratch through the tiny-YOLO architecture built using Keras.

In my understanding,
the labels for each image should be of the form:
[p_c, b_x, b_y, b_h, b_w, c_1, c_2] or variants for other cases.

In this case, there are 7 elements in each label for an image.

Final layer is a convolutional layer with output (3, 3, 7), with each channel representing the label.

However, during training and at the final layer, the linear activation of the convolutional layer expects 4 dimensions whilst the labels I've provided is of the form (m, 7) as explained above. (note: m = number of training data).

What should I do to fix this? Any help will be appreciated!

Below is a model summary and the error at the bottom.

   Layer (type)                 Output Shape              Param #

=================================================================

conv2d_1 (Conv2D)            (None, 100, 100, 16)      448

_________________________________________________________________

activation_1 (Activation)    (None, 100, 100, 16)      0

_________________________________________________________________

max_pooling2d_1 (MaxPooling2 (None, 50, 50, 16)        0

_________________________________________________________________

conv2d_2 (Conv2D)            (None, 50, 50, 32)        4640

_________________________________________________________________

leaky_re_lu_1 (LeakyReLU)    (None, 50, 50, 32)        0

_________________________________________________________________

max_pooling2d_2 (MaxPooling2 (None, 25, 25, 32)        0

_________________________________________________________________

conv2d_3 (Conv2D)            (None, 25, 25, 64)        18496

_________________________________________________________________

activation_2 (Activation)    (None, 25, 25, 64)        0

_________________________________________________________________

max_pooling2d_3 (MaxPooling2 (None, 12, 12, 64)        0

_________________________________________________________________

conv2d_4 (Conv2D)            (None, 12, 12, 128)       73856

_________________________________________________________________

activation_3 (Activation)    (None, 12, 12, 128)       0

_________________________________________________________________

max_pooling2d_4 (MaxPooling2 (None, 6, 6, 128)         0

_________________________________________________________________

conv2d_5 (Conv2D)            (None, 6, 6, 512)         590336

_________________________________________________________________

activation_4 (Activation)    (None, 6, 6, 512)         0

_________________________________________________________________

max_pooling2d_5 (MaxPooling2 (None, 3, 3, 512)         0

_________________________________________________________________

conv2d_6 (Conv2D)            (None, 3, 3, 1024)        4719616

_________________________________________________________________

activation_5 (Activation)    (None, 3, 3, 1024)        0

_________________________________________________________________

conv2d_7 (Conv2D)            (None, 3, 3, 1024)        9438208

_________________________________________________________________

activation_6 (Activation)    (None, 3, 3, 1024)        0

_________________________________________________________________

conv2d_8 (Conv2D)            (None, 3, 3, 7)           7175

_________________________________________________________________

activation_7 (Activation)    (None, 3, 3, 7)           0

=================================================================

Total params: 14,852,775

Trainable params: 14,852,775

Non-trainable params: 0

_________________________________________________________________

Error:

    ValueError: Error when checking target: expected activation_7 to have 4 dimensions, but got array with shape (246, 7)

Model Code:

def build_model(img_rows, img_cols, num_channels):

    model = Sequential()



    input_shape = (img_rows, img_cols, num_channels)



    if K.image_data_format() == "channels_first":

        input_shape = (num_channels, img_rows, img_cols)



    model.add(Conv2D(filters=16, kernel_size=3, padding="same", input_shape=input_shape))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    model.add(MaxPooling2D(pool_size=(2,2), strides=2))



    model.add(Conv2D(filters=32, kernel_size=3, padding="same"))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    model.add(MaxPooling2D(pool_size=(2,2), strides=2))



    model.add(Conv2D(filters=64, kernel_size=3, padding="same"))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    model.add(MaxPooling2D(pool_size=(2,2), strides=2))



    model.add(Conv2D(filters=128, kernel_size=3, padding="same"))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    model.add(MaxPooling2D(pool_size=(2,2), strides=2))



    model.add(Conv2D(filters=512, kernel_size=3, padding="same"))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    model.add(MaxPooling2D(pool_size=(2,2), strides=2))



    model.add(Conv2D(filters=1024, kernel_size=3, padding="same"))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    model.add(Conv2D(filters=1024, kernel_size=3, padding="same"))

    # model.add(LeakyReLU())

    model.add(Activation("relu"))

    #Softmax layer (filters: p_c, b_x, b_y, b_w, b_h, 2 classes)

    model.add(Conv2D(filters=7, kernel_size=1, padding="same"))

    model.add(Activation("linear"))



    return model

Training Code:

def main():    

    # Put training data and labels into variables.

    (train_images, train_labels), (test_images, test_labels) = get_data()



    # Normalize training data variables.

    train_images = train_images.astype("float32")/225.0

    test_images = test_images.astype("float32")/225.0







    # Create tiny-Yolo model

    model = TinyYolo.build_model(100, 100, 3)

    model.summary()

    optimizer = SGD(lr = 0.01)



    model.compile(loss = "categorical_crossentropy", optimizer=optimizer, metrics=["accuracy"])

    model.fit(train_images, train_labels, batch_size=12, epochs=20, verbose=1)

    (loss, accuracy) = model.evaluate(test_images, test_labels, batch_size=12,verbose=1)