cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
raghuram1a
Adventurer
Adventurer
609 Views
Registered: ‎12-18-2018

Neural network model to quantize and compile using dnndk tools

Hi,

Iam working on DEEPHI DNNDK tool kit to run sample image classification examples. AT present I am working 3DCNN generic neural network model . Could anyone kindly suggest me is decent and dnndc compilers will support any neural network model other than resnet50 inception. 

 

Thanks,

Raghuram

0 Kudos
3 Replies
quentonh
Xilinx Employee
Xilinx Employee
596 Views
Registered: ‎05-24-2019

@raghuram1aAbsolutely, there are a number of image classification example networks that are supported.  See page 22 of UG1354:

https://www.xilinx.com/support/documentation/user_guides/ug1354-xilinx-ai-sdk.pdf

Or, you can deploy your own custom network.  Here is a link to the tutorials that are available on Github:

https://github.com/Xilinx/Edge-AI-Platform-Tutorials

--Quenton

0 Kudos
gwjjj35
Observer
Observer
554 Views
Registered: ‎04-16-2019

Can I interpret "I can deploy my own custom network" as "There is no error with dnnc output"?

 

Note:My own model based on VGG,But I made some changes to the structure of the network.For example,the output of the full connectivity layer of the network is 2 classes instead of 100.

Can I deploy my model by DNNDK?

 

Thank you!

微信图片_20190723150351.png
0 Kudos
kjgreenwood
Adventurer
Adventurer
537 Views
Registered: ‎04-10-2019

You can make changes to the structure of models like you did or even create your own. But there are areas where you might need to get creative if your network uses a Caffe layer that isn't supported by DNNC.

0 Kudos