Maximum supported model size?

[idata]

Models can become quite large: what are NCS' limits here e.g. in terms of maximum network neurons/layers?

[idata]

@ncsnon As far as neurons/layers, there really isn't a hard cap for these although there are some size limitations that will limit the model. The first being a maximum graph model size limit of 320 MB. The other being a memory allocation limit which is calculated in the FileIO.py file found in /opt/movidius/NCSDK/ncsdk-x86_64/tk/Controllers directory in the line if zero_data_offset - buffer_size + pad + buffer_index > 100 * 1024 * 1024: throw_error(ErrorTable.NoResources)

[dilip96]

HI, this answer is given for NCS 1. what about NCS2?

What is the model size limitation on Myriad X VPU?

100 * 1024 * 1024 this value in above answer gives ~105 Mb. In that case, what is graph model size limit of 320 MB mean? are we talking about a different model size?

If I am developing a custom a model, how do I set bar on the model size so that I can run inference on VPU without overloading?

[idata]

@Tome_at_Intel

What are these variables zero_data_offset, buffer_size, pad, buffer_index and 100 * 102 * 4 * 1024 means?

How do they correspond to Caffe deploy.prototxt? so as to calculate the Caffe's deploy.prototxt and know if it is within limit.

because I occured with the similar situation that I tried to converted my own module and faced with the error message about resources limit.

[Error 35] Setup Error: Not enough resources on Myriad to process this network.

[idata]

@miily If you reached the resource limit, then the network may be too large to be processed on the NCS device. The variables you listed are used to calculate the approximate size of the network.