Help to bootstrap the project (begineer)

[bbartling]

Hi,

I got some tips from the Intel community message board:

Steps to start a custom person detection model training:

1. • Follow Steps 1 to 5 in training_extention documentation.
2. • Deactivate the virtual environment once completed.
3. • Follow Steps 1 to 5 in person detection documentation.

In step noted above, am I supposed to clone a particular branch? If I clone per the directions in the README:

git clone https://github.com/openvinotoolkit/training_extensions.git
export OTE_DIR=`pwd`/training_extensions

I think it just gives me the develop branch on my Ubuntu 20.04 machine, this is what I am on:

$ git status
On branch develop
Your branch is up to date with 'origin/develop'.

On step 5 further down....on the develop branch I cannot complete this step: pip3 install -e ote/

The develop branch I have a ote_cli directory and ote_sdk directory....will one of these work? If did a pip install -e on both.

Also notice the develop branch doesn't have a models directory...thanks for any tips getting me going

[JihwanEom]

Please use our latest develop branch and refer quick start guide. https://github.com/openvinotoolkit/training_extensions/blob/develop/QUICK_START_GUIDE.md

[bbartling]

Thank you for the response....am very new to openVINO so any beginner level language/tips greatly appreciated.

On the dev branch quick start link that you provided me I have the detection virtual env up and successful up to the find ote step where I run: ote find --root $TASK_ALGO_DIR

But just return empty brackets [] is because I don't have any of the pretrained open model zoo models downloaded?

Can I still re-tune an existing pretrained model in the open_model_zoo library, these steps with the people detection 2020.

I have a data set prepared on my machine of a directory of jpeg images of people which I am hoping to try out.

Also curious on the ote train do I need to come up with my own label_schema.json if I am using my own dataset or is that something already built into the people detection 2020 even if I use my own dataset?

[bbartling]

Regarding the image annotation statement that I asked above (the label_schema.json) I used this tool called labelImg which generates an XML file for each image.

Looks like this:

Just curious if this is something I can use to retrain an existing pretrained open_model_zoo model but could use some help getting started.

[bbartling]

Im getting further help here:

https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/openVINO-extensions-retrain-a-model/m-p/1430808#M28650