DL workbench configuration for faster rcnn object detection gives accuracy 0

[raushan]

DL workbench gives accuracy 0 for my faster rcnn inception v2  coco tensorflow object detection model v 1.15. Please help with the Configuration for Accuracy setting , I tried calculating accuracy for an ssd mobilenet v2 coco model and the accuracy is 86% but 0 in the case of faster rcnn inception v2  coco tensorflow object detection model. 

I have attached the screenshot of the DL workbench

I have gone through this link but still accuracy is 0 https://docs.openvinotoolkit.org/latest/workbench_docs_Workbench_DG_Configure_Accuracy_Settings.html

[Iffa_Intel]

Greetings,

You may try this as a workaround: https://docs.openvinotoolkit.org/latest/workbench_docs_Workbench_DG_Int_8_Quantization.html

This is to Boost a model with INT8 inference and measure accuracy

Sincerely,

Iffa

[raushan]

Hi, 

Thank you for the response.

I tried importing frozen graph .pb file into DL workbench. In first case the override output option was given as 'Softmax' and after optimizing with Int8 precision we got Accuracy 0. In second case the model was imported without any override output and it failed to optimize the model to Int8 format. The error said 'resize to input size possible, only for one input layer case'.

I have attached screenshot also. Please help out in this issue.

[Iffa_Intel]

Hi,

Same as what I replied to your other thread, I really appreciate if you could attach your frozen model graph and also your dataset , so that I could replicate.

Sincerely,

Iffa

[Iffa_Intel]

Greetings,

Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question

Sincerely,

Iffa

[Alexander_D_Intel1]

Hi @raushan ,

We have answered your question in another thread: https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/DL-workbench-accuracy-tool-is-failing/m-p/1210162

If you have any additional questions or proposals, please ask them via the official DL Workbench Discussion Forum.