cancel
Showing results for 
Search instead for 
Did you mean: 
Visitor
Visitor
323 Views
Registered: ‎11-04-2019

Evaluating Tensorflow Resnet50 on Computer

Jump to solution

I'm trying to evaluate the frozen graph for resnet50 using Tensorflow on my computer (Ubuntu 18.04.3). The evaluation itself works fine (using the "evaluate_frozen_graph.sh" script), but I'm getting extremely low testing accuracy, on the order of 0.001 for top1 accuracy. I was wondering if this is possibly due to the labels I'm using. In fact, I'm not sure exactly what labels should be used here. I'm testing my model on the ILSVRC2012 validation set, which doesn't have any labels, at least when downloaded from the suggested site (http://academictorrents.com/collection/imagenet-2012). Can anyone point me to where I might find the labels for this dataset in the format that is expected by the pretrained DNNDK resnet50 model?

0 Kudos
1 Solution

Accepted Solutions
Highlighted
Xilinx Employee
Xilinx Employee
234 Views
Registered: ‎03-27-2013

Re: Evaluating Tensorflow Resnet50 on Computer

Jump to solution

Hi @achempak ,

 

I double check the webpage. It seems that the calibration label file is not attached in that document. Sorry about that.

Actually I don't have a label file for all the images.

This is what I got here: about 4000 labels on ILSVRC2012_val dataset for calibration.

Please check the attachment to get the caibration label file for ILSVRC2012_val dataset.

Hope this can help.

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------

View solution in original post

0 Kudos
4 Replies
Highlighted
Xilinx Employee
Xilinx Employee
277 Views
Registered: ‎03-27-2013

Re: Evaluating Tensorflow Resnet50 on Computer

Jump to solution

Hi @achempak ,

 

I would suggest you to refer to this AR: https://www.xilinx.com/support/answers/72804.html

You can find a lot of helpful information there.

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------
0 Kudos
Highlighted
Visitor
Visitor
249 Views
Registered: ‎11-04-2019

Re: Evaluating Tensorflow Resnet50 on Computer

Jump to solution

Hi @jasonwu , I looked at the link you posted, but I don't think it contains information to solve this problem. I was wondering how the ResNet network was trained (e.g. what were the labels used) such that I can use the same labels for testing. I may have missed something, though!

0 Kudos
Highlighted
Xilinx Employee
Xilinx Employee
235 Views
Registered: ‎03-27-2013

Re: Evaluating Tensorflow Resnet50 on Computer

Jump to solution

Hi @achempak ,

 

I double check the webpage. It seems that the calibration label file is not attached in that document. Sorry about that.

Actually I don't have a label file for all the images.

This is what I got here: about 4000 labels on ILSVRC2012_val dataset for calibration.

Please check the attachment to get the caibration label file for ILSVRC2012_val dataset.

Hope this can help.

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------

View solution in original post

0 Kudos
Highlighted
Visitor
Visitor
212 Views
Registered: ‎11-04-2019

Re: Evaluating Tensorflow Resnet50 on Computer

Jump to solution

Hi @jasonwu,

This is what I was looking for. Thank you!

0 Kudos