Problem with Conv2dtranspose for super resolution

TensorFlow General Discussion
1

Hi
When I doing the deconvolution for super-resolution, the Conv2dtranspose generates a black bar on the top (2xh in pixels) and left (2xw in pixels).

The photo: drive-google-com/file/d/12Ob2gVrd52j2mDbQ8_wMSUDkISmA37ZA/view?usp=sharing

The input before Conv2dtranspose layer is (600,720,1), and it should be 4x times bigger to (2400, 2880,1) after the Conv2dtranspose.
The layer is : out = Conv2DTranspose(num_channels, 9, strides=scale_factor, padding=‘same’)(x)
Does anyone familiar with this?
Thanks.

2

Update, The layer is : out = Conv2DTranspose(1, 9, strides=4, padding=‘same’)(x)