# IndoFashion : Apparel Classification for Indian Ethnic Clothes

**Authors:** Pranjal Singh Rajput, Shivangi Aneja  
**arXiv:** [2104.02830](https://arxiv.org/abs/2104.02830)  
**Date:** 2021-04-06  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2104.02830v1/

## Key points

- Indian ethnic clothing has a diverse range of categories for both men and women, such as "Saree" and "Dhoti".
- Traditional Indian garments have unique styles and patterns distinct from Western attire like t-shirts and jeans.
- Existing models struggle to classify ethnic outfits due to their dissimilarities from Western fashion.
- A large-scale ethnic dataset with over 106,000 images encompassing 15 categories was introduced for fine-grained classification of Indian ethnic clothing.
- The dataset was curated from various Indian e-commerce websites to ensure a diverse representation of ethnic attire.
- An impressive classification accuracy rate of 88.43% was achieved through rigorous evaluation against baseline models.
- The dataset has the potential to catalyze advancements in algorithm development across multiple domains beyond cloth categorization tasks.
- This research opens up possibilities for future innovations that can enhance the accuracy and efficiency of algorithms tailored for classifying Indian ethnic clothes.

## Summary

In the realm of cloth categorization, a crucial research problem utilized by e-commerce platforms to accurately present products to consumers, Indian ethnic clothing stands out with its diverse range of categories for both men and women. The distinctiveness of traditional Indian garments such as "Saree" and "Dhoti" in comparison to Western attire like t-shirts and jeans is evident in their unique styles and patterns. This disparity poses a challenge for existing models trained on standard datasets, which struggle to effectively classify ethnic outfits due to their dissimilarities from Western fashion. To tackle this issue head-on, a groundbreaking large-scale ethnic dataset comprising over 106,000 images encompassing 15 different categories specifically tailored for fine-grained classification of Indian ethnic clothing has been introduced. This dataset was meticulously curated from numerous Indian e-commerce websites, ensuring a diverse representation of ethnic attire. Through rigorous evaluation against various baseline models for cloth classification tasks, an impressive classification accuracy rate of 88.43% was achieved. The significance of this pioneering dataset extends beyond its immediate application in cloth categorization; it holds the potential to catalyze advancements in algorithm development across multiple domains including landmark detection and other specialized tasks related to ethnic clothing. By shedding light on the intricacies of classifying Indian ethnic clothes, this research paves the way for future innovations that can enhance the accuracy and efficiency of algorithms tailored for this unique segment of the fashion industry.