Using .yml in DL workbench

[ps2023]

Hi ,

I am using DL Workbench 2022.1 and want to benchmark the unet-camvid-onnx-0001 ( https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/intel/unet-camvid-onnx-0001) in the DL Workbench.

I am not sure how to import the model in DL Workbench as it not available for download through the Model Zoo option in Workbench. Further the Intel Model Zoo github repo has two files namely  accuracy-check.yml and model.yml file and I am not sure how to use them either.

Can you recommend how to import the unet model to DL workbench available on github repo?

Thanks

[IntelSupport]

Hi Ps2023,

Thanks for reaching out.

To import the unet-camvid-onnx-0001 model into DL Workbench, you may download the model with omz_downloader on your machine. Then import the Intermediate Representation (IR) files into your DL Workbench as below:

Regards,

Aznie

[ps2023]

Thanks Aznie, I was able to download the Unet model. Also does it matter what domain is selected, as in the attached image it says NLP?

[IntelSupport]

Hi Ps2023,

Yes, you have to select the domain based on the framework of the model. Here are the details for each domain.

Computer vision (CV) models are used to extract meaningful data from digital images and act or make predictions based on that information. Computer vision use cases include object detection, classification, segmentation, image inpainting, style transfer, etc.

Natural language processing (NLP) models are used to process and interpret human language through different use cases, for example, text classification, textual entailment, etc.

For unet-camvid-onnx-0001 model, the domain will be CV as the model is designed to perform semantic segmentation.

Regards,

Aznie

[ps2023]

Thanks. I had a follow-up question

The unet-camvid-onnx-0001 model was trained using the CamVid dataset.

1) Can I download the dataset which Intel used it for training?

2) CamVid dataset is not supported by DL Workbench, how can I use it to measure the accuracy of the model using CamVid in DL Workbench?   

[IntelSupport · Accepted solution]

Hi Ps2023,

The dataset is not available to be downloaded. The unet-camvid-onnx-0001 is the per-pixel probabilities of each input pixel of the 12 classes of the CamVid dataset. You can find the RGB value of the classes in this directory:

INSTALL_DIR\deployment_tools\open_model_zoo\data\palettes

Check out this Dataset Types for the list of dataset types available to use in the DL Workbench.

On another note, in the DL workbench, you can create an accuracy report. Refer to this Accuracy Report documentation.

Regards,

Aznie

[IntelSupport]

Hi Ps2023,

This thread will no longer be monitored since we have provided information. If you need any additional information from Intel, please submit a new question.

Regards,

Aznie