A Pitiful Embedded Engineer Look For Help..

[idata]

In caffe,

we can use the following codes to assign different input shape for the network, net_12.blobs['data'].reshape(1,3,ws,hs) net_12.blobs['data'].data[…]=scale_img # net_12 ' s input shape is 1x3x12x12, but neither parameter ws nor hs equals 12.but in NCSDK 2.05, if we put different shapes into the graph,an input length dismatch error would occur.

[idata]

docs for official :

In NCAPI v2 input must be written to a properly initialized and created Fifo that will be passed to the Graph.queue_inference() method for inference.

So it means the tensor we put into the queue should have the same shape as data layer's shape which have been defined in the graph file.

So, if there any solution to implement a different shape tesors into the same graph?

[idata]

E: [ 0] ncFifoSetOption:2815 trying to set host tensor decriptor to a shape that doesn't match graph tensor descriptor shape!

[idata]

@zufeifei The NCS and NCSDK is designed for deployment. As such, any models that are compiled by the NCSDK are considered ready to be deployed and no changes can be made to the model once the compilation happens. If any changes need to be made to the model, a new Movidius graph file must be generated.