How to obtain the output shape in dynamic shape condition

Hi, as far as I’m concerned, in dynamic input shape condition, the output shape is calculated through input shape.

Assume the input shape maybe something like [1, 3, -1, -1], and output shape be like [1, 3, -1, -1].

We can assign input shape with the shape of input data, but the output shape should be calculated through running a forward of your model, so that you can allocate output buffer for gpu, can anybody give me some advice? Thanks in advance!

@NVES_R Any advice?