webcam.py about yolov7_openvino_cpp-python HOT 13 CLOSED

Thanks for your feedback. I have fixed it. Pls try it again

from yolov7_openvino_cpp-python.

Great....thank you its working. Also curious to try the Yolo7 tiny....is that something I can just download a model from and try out with webcam.py?

Also curious if you had the time, is what was upgraded in your code to accommodate a video feed? Was there a lot of effort that went into making the pipeline asnyc?

from yolov7_openvino_cpp-python.

Also playing around with the -b arg...like py -3.7 webcam.py -i 0 -m .\model\yolov7.onnx -b 2 or a -b 4 and I get an error like this:

Requested output shape {1,3,85,20,20} is incompatible with input shape {2, 255, 20, 20}

or for -b 4 like this:
Requested output shape {1,3,85,20,20} is incompatible with input shape {4, 255, 20, 20}

from yolov7_openvino_cpp-python.

Also playing around with the -b arg...like py -3.7 webcam.py -i 0 -m .\model\yolov7.onnx -b 2 or a -b 4 and I get an error like this:

Requested output shape {1,3,85,20,20} is incompatible with input shape {2, 255, 20, 20}

or for -b 4 like this: Requested output shape {1,3,85,20,20} is incompatible with input shape {4, 255, 20, 20}

Sorry I dont see this problem in my system

from yolov7_openvino_cpp-python.

Ill have to play around with this a bit more....would I need to make sure my video frames have the correct shape?

Also curious, I have a NCS2 stick laying around but no GPU....can I run the Yolo7 Tiny on NCS2 & CPU?

from yolov7_openvino_cpp-python.

Ill have to play around with this a bit more....would I need to make sure my video frames have the correct shape?

Also curious, I have a NCS2 stick laying around but no GPU....can I run the Yolo7 Tiny on NCS2 & CPU?

1. No You don't need to correct shape, it will automatically resize the input data.
2. For CPU the answer is sure. For NCS2 probably not, but I have not tested yet. If you have CORE CPU, there is definitely an integrated GPU with CPU. You can try it.

from yolov7_openvino_cpp-python.

Ill have to play around with this a bit more....would I need to make sure my video frames have the correct shape?

Also curious, I have a NCS2 stick laying around but no GPU....can I run the Yolo7 Tiny on NCS2 & CPU?

BTW, If you have further requirement for a project case, maybe we can discuss it and help for optimization.

from yolov7_openvino_cpp-python.

Hi, late response here....and thanks again for your continued help, not a lot of wisdom here : )

I forked your repo and rigged in a Flask App to view the results in the browser remotely with a VPN connection to a remote site. Locally the edge device is just a CPU with a web cam. I use your code to verify the people counting accuracy in a building conference room. I work in research for the smart building industry where I am using computer vision to quantify the results of some other technology that counts people via Wi-Fi access point technology. Ultimately we will test some concept ideas to turn off air conditioning systems to save energy bill and carbon footprints.

What I notice is that running the Yolo 7 model via on OpenVINO I can gain a whole 1.5 plus FPS which is actually nice. The people don't move fast in the conference room so even a .5 FPS actually works well for what I need it for. Ultimately the web app has a rest endpoint where we log the data and view charts made in Grafana to see how well the technology works.

FPS isnt shown below but not running the Yolo 7 model via open VINO frame rate is only .5:

If you had the time could I ever get some more help on the async features on my IoT app? I haven't been able to get that to work. Would you ever be will to try out the IoT app I made? If you had any tips to write better code feel free as well scripting is something I only have a few years experience in.

At some point in time there is a paper to be published, some white paper for Wi-Fi access point technology where in an appendix I post a link to the computer vision that was used to verify Wi-Fi access point technology accuracy.

from yolov7_openvino_cpp-python.

Thanks for your feedback. Glad to see the basic function is working in your conference room.

1. To improve the yolov7's performance, you can try yolov7 tiny or compress your current yolov7 model with OpenVINO's quantization tool (you can try this notebook: https://github.com/openvinotoolkit/openvino_notebooks/blob/develop/notebooks/226-yolov7-optimization/226-yolov7-optimization.ipynb).
2. I have updated code today, and you can pull the latest version for better performance on async infer.
3. What's the issue you meet with your IoT app, maybe I can try it when I am free.

from yolov7_openvino_cpp-python.

Hi Ethan, ill try the Tiny model like you mentioned to see if I can get better performance to start with.

With the IoT app I just havent integrated any of the async pipe line features...but Ill give it a shot : )

Ill close this issue for now and open a new one if I run into a snag with the asnyc implementation inside of the webcam.py. Thanks again for the continued help/support...this has been a fun learning project for me.

from yolov7_openvino_cpp-python.

@OpenVINO-dev-contest would you have any tips for running the notebook file you sent in the link to convert the Yolo v7 to int8?

I also submitted a git issue:
openvinotoolkit/openvino_notebooks#708

from yolov7_openvino_cpp-python.

Hi @bbartling Since it's a development branch, maybe some feature is not ready for release and I think our notebook maintainer team will help to solve this issue, thanks.

from yolov7_openvino_cpp-python.

BTW did you follow this guide to set-up your environment ? https://github.com/openvinotoolkit/openvino_notebooks#-installation-guide

from yolov7_openvino_cpp-python.