25th Sept 2024 - 16th Nov 2024

Challenge Overview

Did you ever go through a product on an e-commerce website where the image is showing a short sleeve shirt while the product description says full-sleeve?

Predicting attributes from supplier uploaded images as part of cataloging is extremely crucial for any E-commerce platform. Typically, when suppliers upload their products for listing on the marketplace, they are required to fill in various details corresponding to product attributes (e.g., color, pattern, sleeve). This process often results in incorrect or incomplete attribute information.

This motivated us to release a subset of Meesho products dataset to facilitate growth in this area in the Indian context.

Dataset

The e-commerce product dataset will contain product information and high-quality images. Each product entry will include the following attributes:

Product ID

Unique identifier for each product.

Product Image

URLs of images that are publicly available online or can be accessed using productid.jpg.

Category

Attribute Keys

The attribute name specified for the product.

Attribute Values

The attribute values specified for the product corresponding to the attribute key.

The rights to this dataset are only for non-commercial use.

License

Evaluation

We are going to evaluate the attribute classification model on a product dataset of the same categories. The attribute identification problem is posed as a multi-class classification problem at the attribute level. In our setting, an attribute can only take one value and hence it’s not a multi-label problem. To evaluate the model fairly, we will use micro and macro f1-score at attribute level. To calculate the f1 score at the category level, we take the unweighted average across all attribute f1 scores for that category.

Expected outcomes from participants

Top 15 candidates on leaderboard will be will be asked to shared the complete code (training & inference both) hosted on the github explaining how to run the code and document explaining your approach.

Google form link

Challenge Timelines

Registration and competition start

Registration and competition entry ends

Last date to make submission on Kaggle

Last date for report submission

Winners will be announced