Originally, this was a challenge about fashion image classification. We have thousands of fashion images, and want to classify them into different classes. Of the classification, it specified 6 categories (or dimensions), such as sleeve, neckline, texture, decoration pattern and so on. And for each category, it has several classes.
Category 1 | Category 2 | Category 3 | Category 4 | Category 5 | Category 6 |
---|---|---|---|---|---|
floral | long_sleeve | maxi_length | crew_neckline | denim | tight |
graphic | short_sleeve | mini_length | v_neckline | chiffon | loose |
striped | sleeveless | no_dress | squared_neckline | cotton | conventional |
embroidered | no_neckline | leather | |||
pleated | faux | ||||
solid | knit | ||||
lattice |
This means, for an individual image, we need to predict 6 class labels, and each class label corresponds to one category.
The dataset has bounding box and landmarks.