Task: Multilabel Classifiation || Data: Images scraped from fashion website https://www.myntra.com/. || Total images: 50200 || Train Images: 32630 || Validation Images: 17570 || Number of Classes: 24 || Classes:
['Activewear', 'Bottomwear', 'Dresses', 'Dungarees', 'Ethnicwear', 'Fullbodywear', 'Innerwear_and_Sleepwear', 'Jackets', 'Jeans', 'Jeggings', 'Jumpsuits_and_Bodysuits', 'Lingerie_and_Sleepwear', 'Pants', 'Shirts', 'Shorts', 'Skirts', 'Suits', 'Swimwear', 'Tops', 'Tshirts', 'Upperwear(inner)', 'Upperwear(outer)', 'female', 'male']
File paths and their corresponding labels are stored in a CSV file named 'File_finale8.csv'
I learnt multi label cassification through fastai lectures 'Practical Deep Learning for Coders' by Jeremy Howard. In the class, Jeremy shows how to apply mutilabel classification to satellite images. I wanted to practice what he taught. But as in the machine learning industry they say, "Never test your model on your training set", rather than reworking the same project I applied my knowledge to a different dataset. This technique was amazing and I learnt a lot more than I would have learnt by just making the same project.
The biggest thing I learned was the ability to read Fastai source code and identify bugs in it. I even wrote my own code to get better results.
Just download the Clothing_Classifier_with_fastai_2.ipynb notebook or click on open in colab. In notebook settings choose GPU. The notebook is well-documented, so just follow the instructions and you will be fine.
Python 3 Fastai Jupyter Notebook
If you are running on Google Colaboratory, you will be fine with python 3. But, you still might have to install Fastai. Here is the link to official fastai setup instructions for colab https://course.fast.ai/start_colab.html