CHaiDNN v2 is now live!

[kamrank]

Hello Everyone! 

I'm excited to announce that CHaiDNN v2 is now available on Github! CHaiDNN is a Xilinx Deep Neural Network library for acceleration of deep neural networks. V2 brings with it many exciting enhancements:

• 3.5x performance improvement from CHaiDNN-v1 on Image Classification Benchmarks
• 4x effective GOPS improvement over CHaiDNN-V1
• Enhanced documentation, examples and tutorials.
• Support for software custom layer plug-ins
• Ability to segment networks into subgraphs, and partition between hardware/software
• ModelZoo of 6 networks at int8 and int6 precision
• Support for URAM devices
• Scalable to be implemented in small to large Zynq MPSoC devices.
• Introducing DIET_CHAI for Image Classification with Googlenet v1 and ResNet
• A miniature version of CHai for smaller MPSoC/ Zynq devices
• Try Pre-Built Binaries to get started today!

Full list enhancements here.

Want to learn more about CHaiDNN and some of the new features? Stay tuned to the SDSoC Forum. We will be posting tech blogs, which will go into detail on these new features and allow you to ask your questions directly to CHaiDNN's creators! 

Thanks,

Kamran 

[diego73]

This is awesome, I'll try it on my ZCU104 as soon as I find a chance!

[xiaohong_liut]

Do you know where can I find model training source code? I found SSD caffe model source code. but the prototxt architectures are not same. 

Also, how to generate "data" folder? 

Thanks, Lily

[aeeshms1]

Hi,

How can I compile the CHaiDNN for sdaccel flow?

[kamrank]

Currently, the examples for chai are only available for sdsoc. for sdaccel flows you can also take a look at ml suite: https://github.com/Xilinx/ml-suite