Faster RCNN change backbone

Hello guys

I am trying to change the RESNET50 backbone of Faster RCNN by MobileNET.
My code seems like:

from torchvision.models.detection import FasterRCNN
backbone = torchvision.models.mobilenet_v2(pretrained=True)
backbone.out_channels = 1280
model = FasterRCNN(backbone)

But i get this error

ValueError: num_classes should not be None when box_predictor is not specified

I am starting with this kind of networks, so there are many things that i am still learning, any help would be great.
The objective that i have is to compare the time inference of resnet 50+faster (default) vs mobilenet+faster, so i dont want either train or fine tuning the network

See how fasterrcnn_resnet50_fpn use FasterRCNN

def fasterrcnn_resnet50_fpn(pretrained=False, progress=True,
                            num_classes=91, pretrained_backbone=True, **kwargs):
    if pretrained:
        pretrained_backbone = False
    backbone = resnet_fpn_backbone('resnet50', pretrained_backbone)
    model = FasterRCNN(backbone, num_classes, **kwargs)
    if pretrained:
        state_dict = load_state_dict_from_url(model_urls['fasterrcnn_resnet50_fpn_coco'],
    return model
# FasterRCNN __init__ function
if num_classes is not None:
    if box_predictor is not None:
        raise ValueError("num_classes should be None when box_predictor is specified")
        if box_predictor is None:
            raise ValueError("num_classes should not be None when box_predictor "
                                 "is not specified")

Specify num_classes or box_predictor.

I added this

model = FasterRCNN(backbone, num_classes=2) # Aircraft + background

pred = model(img)

And this is what i get

RuntimeError: Expected 4-dimensional input for 4-dimensional weight 1280 1280 3 3, but got 2-dimensional input of size [1, 1000] instead

img has the next shape : (1, 3, 920, 1080)