Methods: Our Development
Step 1: Coding
We coded Aurora in Python implementing the most recent image segmentation neural network practices. We intentionally coded the neural network at every layer and every neuron to specialize to our specific task.
Step 2: Images
We collected a database of images of diverse human faces, paying special attention to ensure the presence of gender and racial diversity. These images were then annotated by key opinion leaders from different entities in the aesthetic industry with three depths of injections in mind: deep correction, medium correction, and superficial injection.
Step 3: Training
We ran the images with their corresponding annotations were run through the code to "teach" the code where it has to recommend injections: this is deep learning. Hundreds of epochs and hours of training guided by accuracy and precision analysis at each step drove constant code improvements. We consistently tested the algorithm on unseen KOL-annotated images to gauge its performance.
Step 4: Finalization
After identifying optimal hyperparametrs of the neural network to maximize result accuracy, we integrated the completed code into the Aurora app!
