Garment Classifier

This repository contain an implementation of a Convolutional Neural Networks for classifying garments together with a data scraper script, used to download images from Zalando's online catalogue.

See in depth tutorial on Medium Automatic Classification of an online Fashion Catalogue: The Simple Way

Scraper

--base_folder         Base folder for downloaded images
--n_images            Number of images to download 
--threads             Number of threads to use

Run Scraper

$ python main.py --base_folder='Data/'  --n_images=100 --threads=10

Training

Once the data has been downloaded, the training phase can start. The training file is train.py and receives the following parameters:

-- learning_rate      Initial learning rate
--epochs              Number of training epochs 
--batch_size          Minibatch training size
--validation_after_n  Number of epochs before apply validation testing
--logdir              Log folder
--save_after_n        Number of epochs before saving network
--model_description   Model name

Run Train

$ python train.py --batch_size=512 ---epochs=100 --save_after_n=20 --validation_after_n=5

Evaluation

After training your model with the MNIST fashion data you can check how is it performing with the data you have downloaded.

--model_dir           Saved model directory
--model_description   Model name
--data_path           Images folder path
--batch_size          Minibatch size

Run Evaluation

$ python eval.py --batch_size=5 --model_dir='Models/' --model_description='garment_classifier' --data_path='Data/Images'
 --batch_size=5