Decompose convolution of window size 5 by convolution filter window size 3

1

Is it okay to use convolution filter of window size 3 twice instead of using convolution filter of window size 5 once? I think use convolution filter of window size 3 twice require less number of parameters…
(3x3)x2. = 18 < (5x5) =25

Below is example code.

model = nn.Sequential( nn.Conv2d(in_dim, out_dim, kernel_size =5, stride =1, padding=2), nn.BatchNorm2d(out_dim), nn.ReLU())

and

model = nn.Sequential( nn.Conv2d(in_dim, out_dim, kernel_size =3, stride =1, padding=1), nn.Conv2d(in_dim,out_dim, kernel_size=3, stride =1 ,padding=1), nn.BatchNorm2d(out_dim), nn.ReLU())