Face Detection

General

• When making your training data, make sure to choose a good number of samples. Something like 400-800 is good.
• Make sure that there is an equal number of samples for each class (or person that the model will try to recognize).

Usage

1. Run the create_data.py file to create training data.

• Sample usage is: python3 create_data.py --person="{person_name}" --samples="{number_of_samples}"
• This will create a folder named {person_name} in the current directory.
• Do this for each person you want to train the model to recognize.

2. Create a data folder in the current directory.

3. Move the folders containing the training data to the data folder.

4. Run the learn.py file.

• You can run this file on your computer but you might not have a capable enough GPU to run the script at reasonable speeds. Thus, you can use the service Google Colaboratory to run the script. Refer to the Google Colab section for more information.

5. Run the detect_image.py file.

• Sample usage is: python3 detect_image.py --path {image_path}

6. Run the detect_video.py file.

• Sample usage is: python3 detect_video.py --person="{person_name}"

Google Colab

To run the script on Google Colab, first create a folder in your Google Drive where you will create the script and house the training data. Next create a folder in your Google drive called data and move the folders containing the training data to this folder. It'll take some time to upload, but from there you are ready to run the learn.py script. In the main directory within Google Colab, create a new Google Colaboratory notebook and paste the learn.py script into it. Comment out the section that says # Google Colab and comment the section that says # Local. Change your runtime type to GPU and run the script. This will produce two files in the Google Colab directory called detector and labels.json. Download these two files into the local directory where you created the training data, and you are set to run the detect_image.py script.