This project aims to develop an artificial intelligence-based system that allows users to easily identify their face shape and receive eyewear recommendations suitable for this face shape.
-
Model Training:
- Using the EfficientNet-B4 (CNN architecture) model, training was performed in 5 different face shape classes. (Heart, Oblong, Oval, Round, Square.)
- In training the model, the dataset contains 800 samples for each class. And the data set contains a total of 5000 samples.
- As a result of the training, the model reached 84% accuracy.
-
Image Processing:
- Images are processed using the PyTorch and Torchvision libraries.
- The images are subjected to OpenCV's Haar Cascade method for face detection.
- The input data for face shape classification is provided in a size and normalised format suitable for the model.
- A data set suitable for the project was selected in Kaggle.
- Dataset Link