ResUNet doesn't match paper's parameters

[jchen42703]

Author said it's supposed to only have 2.2M parameters. This implementation's ResUNet has 2.4M with:

    model = ResUNet(input_channels=5, base_num_features=16, num_classes=3,
                    num_pool=4, max_num_features=256)
    # calculating # of parameters
    total = sum(p.numel() for p in model.parameters())
    trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)
    print(f"Total # of Params: {total}\nTrainable params: {trainable}")

Not sure why. Probably doesn't have to do with the biases. Might be due to the parameters in 1 single extra expensive conv.
#2