Keras plot_model input and output shape only question mark

Question

I'm trying to code YOLOV1 in Keras, and I use the VGG16 which accepts (224,224,3) while the input is (448, 448, 3) (that's the size used by the authors) so I added a Conv2D and a MaxPooling layer to convert 448 to 224. However, when I use plot_model the input and output shapes of Conv and MaxPooling layers in the middle are all question mark (?).

Here is my code:

   base_model = VGG16(
        input_shape=backbone_img_shape,
        # input_shape=img_shape,
        include_top=False,
        weights='imagenet')
    # pdb.set_trace()
    img_input = Input(shape=img_shape)
    arch = layers.Conv2D(64, 3, padding='same', activation='relu') (img_input)
    arch = layers.MaxPooling2D() (arch)
    for i, layer in enumerate(base_model.layers[2:]):
        arch = layer(arch)
    # arch = base_model(img_input)
    arch = layers.Flatten() (arch)
    arch = layers.Dense(4096, activation='relu') (arch)
    arch = layers.Dropout(0.5) (arch)
    arch = layers.Dense((grid_num_per_axis**2)*(num_bbox_per_grid*5 + n_cls)) (arch)
    arch = layers.Reshape(
        (grid_num_per_axis, grid_num_per_axis, num_bbox_per_grid*5 + n_cls)) (arch)
    model = Model(inputs=img_input, outputs=arch)
    model.summary()
    # pdb.set_trace()
    model.compile(
        # optimizer=SGD(lr=0.001, momentum=0.9),
        optimizer='adam',
        loss=yolov1_loss)
    tf.keras.utils.plot_model(model, to_file='haha.png', show_shapes=True)

where backbone_img_shape=(224,224,3) and img_shape=(448,448,3). I think there is something wrong because, for example max_pooling2d layer had (?, 224, 224, 64) size but the later layers should have had something similar instead they had (?). Please help me, thank you very much.

You should ask a question directly, to me its not clear what you want help with. — Dr. Snoopy, Jul 13 '20 at 15:57
@Dr.Snoopy I think that the question marks show that something is wrong with my model initialization — Dang Manh Truong, Jul 13 '20 at 16:20
No, the question marks are about variable size in that dimension, I do not see anything wrong (and you did not ask a question) — Dr. Snoopy, Jul 13 '20 at 16:21
@Dr.Snoopy I think there is something wrong because, for example max_pooling2d layer had (?, 224, 224, 64) size but the later layers should have had something similar instead they had (?) — Dang Manh Truong, Jul 13 '20 at 16:26
Its beter to use model.summary() to see the structure of your model — Dr. Snoopy, Jul 13 '20 at 17:36

Keras plot_model input and output shape only question mark

0 Answers0