DevCloud visualize fails with error

[PRanganathan]

Hi Experts

I am trying to perform object classification using Yolov4-darket, following are the steps performed

1. Anotated the image using OpenCVAT

2. Converted the image anotation to Pascal format

3. Downloaded the Yolov4-darknet model and the weights

4. Loaded the image in google colab [modified obj.names, obj,data,obj.cfg files], model and performed training and new weights were generated

5. tested the image using training set and it worked

6. converted the darknet to tensorflow2 format using the openvino guidelines  for the Downloaded new weights, config, .names, .data files

7.invoked the model optimizer to generate the IR files [.xml, .bin]

8. performed benchmarking usng benchmark app and got a good accuracy at 40fps

9. wanted to perform inference in Devcloud and got report on inference

10. when i wanted to visualize output in devcloud it failed with  "operands could not be broadcast together with shapes (8112,) (648960,) (8112,)"  

can you pl. advise how to resolve the issue in step 10?

11. POst that i need to create a deployable for target Intel HW

warm rgds

PRanganatha

[AthiraM_Intel]

Hi,

Thank you for posting in Intel Communities.

Could you please let us know which DevCloud you are using?(DevCloud for oneAPI/ Edge/ FPGA)

Thanks

[PRanganathan]

Hi

Following are the links in sequence:

1. Dev Cloud: https://www.intel.com/content/www/us/en/developer/tools/devcloud/overview.html

2. Work With Intel OpenVINO : https://www.intel.com/content/www/us/en/developer/tools/devcloud/edge/overview.html

3. Build and Optimize -> Deep learning Workbench [https://notebooks.edge.devcloud.intel.com/user/u158442/lab/workspaces/auto-Z/tree/Reference-samples/iot-devcloud/dl-workbench/DLWorkbenchLauncher.ipynb?reset ]

4. Launch DL workbench , i was trying to test in NuC11 = i7 CPU[currently DLworkbench is down]

Thank you

rgds

[JesusE_Intel]

Hi PRanganatha,

Could you share your model in Tensorflow 2 format and model optimizer command used to convert to IR? If the model cannot be shared, could you share a link to the base model used for training? Could you also attach a screenshot of the error message? I just tested DL Workbench with YoloV4-TF model from the open model zoo and didn't run into any issues so this may be specific to your model.

Regards,

Jesus

[PRanganathan]

Hi

Attaching the original model, saved model, weights, config file, IR converted set 

[JesusE_Intel · Accepted solution]

Hi Pranganatha,

There may be an issue on how the model was converted from DarkNet to Tensorflow to IR. I was able to convert your model using the steps below and visualize output without any issues on DL Workbench. Could you give that a try and let me know the outcome?

git clone https://github.com/david8862/keras-YOLOv3-model-set

mkdir saved_model

python keras-YOLOv3-model-set\tools\model_converter\convert.py yolov4.cfg yolov4.weights saved_model

mo --saved_model_dir saved_model --input_shape [1,608,608,3] --model_name yolov4

These instructions were taken from here. Also, when importing the IR model to DL Workbench configure the models Inputs as NHWC for the Original Layout option.

Regards,

Jesus

[JesusE_Intel]

If you need any additional information, please submit a new question as this thread will no longer be monitored.