Dose openvino IR support multi-shape images as input ？

[D__eric]

I got an error when i was try to convert a Fully Convolutional Network without fix the input images size, so i was wondering if IR support  multi-shape images as input during inference process。

A resized image as the input is not the way i want，how can i convert a FCN like network to openvino IR?

[Roy_A_Intel · Accepted solution]

Hi Eric

You can specify an arbitrary input size as required, then during the inferencing to work with multi-shape images, use the shape inference feature of the inference engine to work with differently-sized inputs. More on shape inference is available here  https://docs.openvinotoolkit.org/latest/_docs_IE_DG_ShapeInference.html 

Regards

Roy