Faster RCNN Model is not yielding good results on DevCloud Edge Devices

[Bhargavi_Ch_Intel]

Hi Team,

I have tried running the OMZ Object Detection Python demo using Faster RCNN model on server and Devcloud edge platforms.

Please find the reference link for sample: https://docs.openvino.ai/latest/omz_demos_object_detection_demo_python.html#doxid-omz-demos-object-detection-demo-python 

Neural Network Topology: Faster RCNN Inception 

Please find the results below for :

On Server - Saphire Rapids Gold(6438N - BF16) and Platinum(8480 - BF16) : Able to get Throughput of 50+ FPS on CPU Plugin.

On Server - IceLake (8380 - FP32 precision) - able to get 35 FPS.

On Devcloud Edge devices: Not yielding good results on both CPU and GPU.

able to get only the below results for FP32 precision:

1. 6-10 FPS on Intel i7 & i9 & Intel flex-170 on CPU.

2. 7-8 FPS on Intel i7 & i9 and 23 FPS on Intel GPU Flex 170 on GPU.

GPU & CPU looks very low in throughput. Request you to help us on this issue.

Also not able to get good results with Int8, FP16 precisions as well. 

For Intel GPU FLex 170, throughput is around 14-18 FPS on GPU plugin.

Request you to help us on this.

[JesusE_Intel]

Hi Bhargavi_Ch_Intel,

Thanks for posting your issue on the forums! I can definitely help take a look, could you please share the model? I will send you a private message.

Regards,

Jesus

[JesusE_Intel · Accepted solution]

Hi Bhargavi_Ch_Intel,

Could you share some details about the model? This will help me reproduce the issue in case you are unable to provide the model.

• Framework
• Model Architecture
• input type
• input size/shape

Regards,

Jesus

[JesusE_Intel]

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