I Built My Own Fashion Dataset For Deep Learning

52,000 labeled fashion images from 10 categories

Amir Ali Hashemi
3 min readJun 29, 2022
Image by author

Link to the dataset (including the scripts used)

When it comes to computer vision, there are two famous image datasets; handwritten digits-MNIST and fashion-MNIST. The latter dataset is provided by Zalando, one of the largest online clothing retailers, and consists of images of ten different clothing categories.

The main issue with the fashion-MNIST dataset is that it is widely acknowledged to be way too easy, overused, and unable to reflect modern computer vision challenges.

With that being the case, I decided to take advantage of this opportunity and build my own fashion dataset similar to fashion-MNIST but slightly more advanced. I wrote a web scraper script that crawled www.zalando.com and extracted the URL of images from 10 different categories as written below.

CATEGORIES = [Jacket, Pants, Jeans, Shorts, T-shirt, Pullover, Bag, Cap, Sandal, Skirt]

The URLs of the images were later converted into 28 by 28-pixel images through another script. The conversion process was the most time-consuming part of this project and took 7 days to complete. Here you can see a few examples of the dataset:



Amir Ali Hashemi

I'm an AI student who attempts to find simple explanations for questions and share them with others