Clothes Classification using CNN

Purpose

After created a Slipper Classifier using Logistic Regression, I wanted to use a more advanced Machine Learning method to train a more complex model. Expending from slippers, I planed to train a model to recognize different clothes. The database includes 6 categories: hat, pants, shoes, skirt, Tshirt and others.With a total number of 3293 Training images and 366 Validation images. The performance of this model is: Training Accuracy is 96.88%; Validation Accuracy is 88.36%.

Steps

1. Download images from Google

Download images using a Firefox extension tool named "Download all images".

2. Use Data Augmentation to get more pictures

Use Keras image augmentation tool to generate more pictures by mirroring or zooming the existing pictures.

 

3. Developing CNN model

Results

The Training Accuracy is: 96.85%; the Validating Accuracy is: 88.36%.

Others

Please visit GitHub for all database and codes: Project Link