Looks like need some args(input size) for IR with dynamic input shape

[junxnone]

[ INFO ] GFLOPs: -53.7485
[ INFO ] GIOPs: 0.0000
[ INFO ] MParams: 330.5959
[ INFO ] Sparsity: 0.0011%
[ INFO ] Minimum memory consumption: 3.0188
[ INFO ] Maximum memory consumption: 171.4634
...

[etienne87]

i agree, how come this does not take any input size for convolutional models?