cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
axt7568
Visitor
Visitor
822 Views
Registered: ‎07-05-2019

Quantized Yolov3 on Ultra96 using DNNDK

Hello Sir/Ma'am, 

I was able to successfully convert and quantize my Yolov3 model and generate the dpu.elf file and to do this I used the latest version of Decent tool (from DNNDK v3.1), but since this particular project repo(https://github.com/Xilinx/Edge-AI-Platform-Tutorials/tree/master/docs/Darknet-Caffe-Conversion) was made for the initial version of DNNDK(v2.08), I had to use the compiler from that version(DNNC from DNNDK v2.08) to compile and generate the dpu.elf file. Now the problem is that when I try to run it on my Ultra96 board, I get an error complaining about the different versions which make sense. So to get around this issue I had to either use both tools from the same version. Since DNNDK v2.08 doesn't have the board files for Ultra96 (only present for ZCU102), I was wondering as to how I can generate the dpu.elf file for the Yolov3 network using the latest DNNDK version 3.1. I have no issues with the Decent tool but the compiler (DNNC)  has been modified significantly for the two versions wherein the --dpu option has been modified with the --dcf option which needs stuff from the additional tool - DLet not compatible with this project. Thus I would really appreciate it if you could provide me with any solution to this problem. 

 

Sincerely, 

AJ

 

0 Kudos
Reply
1 Reply
wutzxt
Visitor
Visitor
275 Views
Registered: ‎08-31-2020

dcf file can be generated by "dlet -f xxxx.hwh" (run in the Vitis-AI CPU/GPU docker image), the hwh file can be located in your hw project.

0 Kudos
Reply