I have the following issue. I have a trained Yolov3 model which I want to convert into openvino IR format. I want to create 2 version of the model one is FP32 and the other is FP16. I followed the link below for the conversion and successfully converted the model.
In order to get the FP16 model, I provided --data_type FP16 argument while executing the mo_tf.py file. This resulted in the successful creation of the FP16 model file. This was all done on Ubuntu 18.04 platform.
Now I am testing the model on Windows 10 platform with Intel HD Graphics 620 with the .bin and .xml files I obtained after conversion. I am successfully able to run the object_detection_demo_yolov3_async for FP32 model on both -d CPU and -d GPU. However, FP16 model is throwing some error like below:
InferenceEngine: API version ............ 1.6 Build .................. 23780 [ INFO ] Parsing input parameters [ INFO ] Reading input [ INFO ] Loading plugin API version ............ 1.6 Build .................. 23780 Description ....... clDNNPlugin [ INFO ] Loading network files [ INFO ] Batch size is forced to 1. [ INFO ] Checking that the inputs are as the demo expects [ INFO ] Checking that the outputs are as the demo expects [ INFO ] Loading model to the plugin [ ERROR ] program creation failed: Program build failed:
I am executing the following command :
C:\Users\<user_name>\Documents\Intel\OpenVINO\inference_engine_samples_build\intel64\Release>object_detection_demo_yolov3_async.exe -i C:\Users\<user_name>\Desktop\yolov3_fp16_intermediate_xtralis_models/out.mp4 -m C:\Users\<user_name>\Desktop\yolov3_fp16_intermediate_xtralis_models/frozen_darknet_yolov3_model.xml -d GPU
It would be great if someone could help me fix it.
Dear Vora, Aditya,
It seems like you did everything correctly. Any day now we are about to release OpenVino 2019 R2. Likely the issue you are reporting (yolo v3 failing on GPU FP16) will be fixed in R2.
Could you kindly wait for the next release ?
Would you happen to have a MYRIAD ? Maybe you don't. That's something you can try in the meantime with FP16.
Thanks for your patience,
Thanks for your reply. Yes, I shall wait for the next openvino release. I hope I can make it work with the new release. However, I haven't tried with MYRIAD yet. So meantime will try it with MYRIAD if it works. However, in the openvino docs, it has been mentioned that FP16 models can work with intel GPU as well. So just tried with that as it was the easiest option.