Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

VGG16- SSD 300 Running time

Chakraborty__Subhasi
915 Views

I am getting a running time of 180-200 ms on object_detection_sample_ssd.exe on openvino in windows with API version 1.1 and build version win_20180511 while running in CPU. Is this running time correct, as i went through a number of articles online where the running time of ssd 300 was mentioned around 40 ms. 

0 Kudos
7 Replies
Mark_L_Intel1
Moderator
915 Views

Hi Subhasis,

Sorry for the late response,

 

I am trying to look at your question and I want to reproduce what you saw first. But I realized you might missed some information to reproduce the issue. Could you answer following questions or give me some hints?

What the topology/model you were using?

I also want to know the processor name of your system?

When you talked about the article where you looked at the runtime number, which article did you checked, could you give me a link?

Mark

0 Kudos
Chakraborty__Subhasi
915 Views

I was running VGG16- SSD 300 on my pc with openvino and i found that the inference time of the model on a single image is 180-200 ms on CPU. My cpu is Intel(R) Core(TM) i7-7700 CPU @ 3.60GHz with Broadwell micro-architecture. I found that SSD 300 with VGG model gives around 46 fps as quoted in the paper SSD: Single Shot MultiBox Detector with Titan X GPU and batch size 1 and  Intel Xeon E5-2667v3@3.20GHz CPU. I wanted to know the expected running time on Openvino through Intel CPU and GPU.

0 Kudos
Mark_L_Intel1
Moderator
915 Views

Hi Subhasis,

​Thanks for the quick response, I will try to reproduce this and I want to know where you download the model files.

The other question, according to our spec, i7-7700 is KabyLake framework which is 2 generations advance than Broadwell, is there a mistake?

https://ark.intel.com/products/97185/Intel-Core-i7-7700HQ-Processor-6M-Cache-up-to-3_80-GHz

Mark

0 Kudos
center__Chen
Beginner
915 Views

Does this question was be solution ?

I ment the same question as Subhasis

 

I use 「SSD: Single Shot MultiBox Detector」

from  following URL https://github.com/weiliu89/caffe/tree/ssd

And I use

SSD_300x300 in openVINO FPS only has about 5~6 

SSD_512x512 in openVINO FPS only has about 1~2

openVIO FPS value is a lot worse than the paper (SSD)

 

My cpu is  intel(R) xeon(R) cpu e3-1505m v6 @ 3.00ghz

 

Looking forward to your reply , thanks
---

Center

0 Kudos
Shubha_R_Intel
Employee
915 Views

Dear everyone,

Please try the benchmark app on the latest and greatest OpenVino 2019 R2 which was just released last week. Is the problem fixed ?

Let me know.

Thanks for your patience !

Shubha

0 Kudos
center__Chen
Beginner
915 Views

Hi Shubha R. (Intel)

I am sorry to reply you so late

I follow you suggestion use benchmark app to test

I get following result 

 

Count:364 iterations

Duration:61221.7ms

Latency:674.689ms

Throughput:5.94561FPS

 

-------------------following is my test detail -------------------------------

1、use model :SSD: Single Shot MultiBox Detector

2、IR model : use mo.py to convert by myself

3、intel CPU : intel(R) xeon(R) cpu e3-1505m v6 @ 3.00ghz

4、VPU:Mustang-v1000MX8

--------------------------------------------------

What I want to know is 

「is this FPS is normal on openVINO R2」? 

0 Kudos
Shubha_R_Intel
Employee
915 Views

Dear center, Chen,

Are you concerned that it's slow ? 5.9 FPS does seem kind of slow but it may be normal. Have you tried other models besides SSD ? 

Thanks,

Shubha

 

0 Kudos
Reply