This is how we can build end-to-end face recognition app using open source tools.
Current project consist of 3 parts:
[x] 1. Build / Train ML model
[x] 2. Creating web app
[x] 3. Deploy app
First of all it’s necessary to report about data quality with EDA. You can reach the results in ___'s notebook. After this, it will be advisable to set up a route for data preprocessing and converting images into features. This made it possible to provide an optimal model architecture for the task of face recognition. Next, the model was built into the pipeline to capture the stages and prepare for creating the application. And finally, Flask API based app was built and deployed.
So, to bring you explicitness of the contest, the main purpose of the project is to detect a person, identify gender and age, emotions (optional)
The source of images is a well known IMDB-WIKI**–500k+ face images with age and gender labels**, published in 2015. And, to be more precise is the "Faces only (1 GB)" version.
- Detecting with Haar Cascade CLF (the
*.xlmpattern brought by OpenCV’s FrontalFace_Deafault) - Cropping and alignment
- Labeling
- Gray scaling
- Cut off less than 60 dim’s images (according to EDA, provided by notebook’s)
- Resizing (100x100)
Please, follow the Stanford’s Eigenimages course material to get known the idea.
- Extract eigen features
- Dimensional reduction
We’ll go the convenient way by creating a flask application, that appears in such interface:
- Home page
Start with a simple options.
- Application page
Go towards prediction, specify the input format.
- Gender prediction page
Get the output.
