cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
508 Views
Registered: ‎06-30-2019

dnnc "shitf_cut >= 0" failed when compile ssd mobilenet v2 converted from original tensorflow model

Environment

cuda: 8.0

cudnn: 7.0

caffe: chuanqi's  fork (with relu6 implementation)

dnndk: 20190624 version 
decent: 1.2.5

dnnc: 2.05 

dpu target: 1.4.0.1

Problem

I downloaded ssd mobilenet v2 from tensorflow's model zoo and converted it to caffe model(float.caffemodel and float.prototxt) using tool provided by chuanqi. "decent.sh" excuted successfully and generated deploy.caffemodel and deploy.prototxt(provided in txt format) I encounter this problem when ran "dnnc.sh":

Screenshot from 2019-07-11 11-14-36.png

Do you guys have any idea about this error? Thanks in advance.

Best regards

0 Kudos
Reply
1 Reply
jcory
Xilinx Employee
Xilinx Employee
473 Views
Registered: ‎02-17-2011

Hi @chuanliang.xie@gmail.com,
Please see the following post as a response to a similar problem:
https://forums.xilinx.com/t5/Deephi-DNNDK/An-error-about-DNNC/m-p/995339#M1510
I'm not sure what the equivalent approach would be in TF to combine BN+Scale layers, but it may be that a similar solution would apply if the root cause of the problem is the same.
Thanks and hope this helps,
-JC
0 Kudos
Reply