Super-resolution not completely respecting input low resolution image

[AhmedGamal411]

Super-resolution not completely respecting input low resolution image

Source: https://www.robots.ox.ac.uk/~vgg/data/voxceleb/vox2.html

Note: There are other instances where it does follow the low input resolution image quite well.

unet0 = NullUnet()  # add a placeholder "null" unet for the base unet

unet1 = Unet(
    dim = unet_dim,
    cond_dim = 512,
    dim_mults = (1, 2, 4, 8),
    num_resnet_blocks = 3,
    layer_attns = (False, True, True, True),
    layer_cross_attns = (False, True, True, True)
)

unet2 = Unet(
    dim = 128,
    cond_dim = 512,
    dim_mults = (1, 2, 4, 8),
    num_resnet_blocks = (2, 4, 8, 8),
    layer_attns = (False, False, False, True),
    layer_cross_attns = (False, False, False, True)
)


imagen = Imagen(
    unets = (unet0,unet1, unet2),
    image_sizes = (8,64, 128),
    timesteps = 1000,
    cond_drop_prob = 0.1
).cuda()`

This images are sampled from the first super-resolution UNet only.

Is there any way to make it follow the low resolution image more closely, i.e parameter tuning ?