Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

Variable input shape

Kranauskas__Justas
1,963 Views

Hello,

Is it possible to use Intel Inference Engine with variable input shape? In Caffe, I have a fully convolutional network which can accept images of practically any size as input but I haven't found how to properly apply Inference Engine without regenerating and reloading network for each possible pair of image (width, height). That's obviously not an option because of the speed issues.

Thanks.

0 Kudos
1 Solution
Nikolay_L_Intel1
Employee
1,961 Views

Hi Justas!

It's posisble to use Inference Engine with variable input size starting from 2018 R3 release of OpenVINO™ toolkit.
Please refer to the "Shape Inference" feature, which allows to change the model input size after reading IR and before loading ICNNNetwork to a plugin. 

Best regards, 
Nikolay

View solution in original post

0 Kudos
2 Replies
Severine_H_Intel
Employee
1,960 Views

Dear Justas, 

the IE cannot handle variable input shapes, the height and width must be defined during the Model Optimizer step. Just the batch size can be dynamically changed for certain topologies at the IE stage.

Best, 

Severine

0 Kudos
Nikolay_L_Intel1
Employee
1,962 Views

Hi Justas!

It's posisble to use Inference Engine with variable input size starting from 2018 R3 release of OpenVINO™ toolkit.
Please refer to the "Shape Inference" feature, which allows to change the model input size after reading IR and before loading ICNNNetwork to a plugin. 

Best regards, 
Nikolay

0 Kudos
Reply