Clothes Classification
The beauty of Data Science is the flexibility in its application.
It can be applied almost to every field: sport, medicine, finance, and fashion too.
This project was entrusted to me with the purpose to develop a program able
to recognize clothes. Working with images is always fun because it allows you to
differentiate from the usual projects.
I received a small dataset containing images belonging to nine clothing categories so
I was facing a multiclass classification problem. I decided to use Transfer Learning
to solve the problem by making use of ResNet a pre-trained neural network fit for images.
Since the number of images per class wasn’t great, I decided to apply Data Augmentation
to expand the training input for the model.
Thanks to some regularization techniques such as dropout I avoided overfitting which is
a common problem to face.
Since the class distribution was unbalanced, I evaluated the model with recall,
precision, and their harmonic mean, F1-score.
This problem task compared to the others was quite easy and quick. I faced similar
problems in my university years, so I was able to manage images and Convolutional Neural Networks.
Unfortunately, I can’t post this project's code or additional information because of
customer policy.