BanglaOrna: A Smartphone Image Dataset of Traditional Bangladeshi Ornaments and Accessories

Published: 5 June 2025| Version 2 | DOI: 10.17632/5c3dykr6pt.2
Contributors:
Shadman Sakib Ahmed, Sabikun Nahar Usmita Ushmi, Gazi Fahim

Description

Traditional Bangladeshi ornaments and accessories—such as churi (terracotta bangles), tikli (filigree forehead ornaments), jhumka (bell-shaped earrings), and Komor Bicha (intricate waist chains)—are emblematic of the nation’s cultural heritage. Despite their anthropological and artistic value, no standardized dataset exists for computational research, hindering applications in heritage preservation, fashion technology, and ethnography. We present BanglaOrna, the first comprehensive dataset with a total of 7,174 high-resolution smartphone images documenting 27 categories of women’s traditional adornments, including kaaner duul (dangling earrings), haar (layered necklaces), and nakphul (bridal nose pins). Collected from Dhaka’s historic markets (New Market, Gausia Market) under diverse lighting and angles, the dataset reflects real-world variability in ornament styles and materials. Images were captured using Redmi Note 14 Pro (200 MP), Redmi Note 14 (108 MP), and Redmi Note 12 Pro+ (200 MP), ensuring fine-grained detail for small, intricate designs. Each image is annotated in COCO format with bounding boxes. Augmentation techniques (Horizontal and Vertical flip (50%), (±10 °) Horizontal and Vertical Shear,Brightness (±25%), Noise ( 0.5% )) were applied to improve model robustness. BanglaOrna opens doors to:  Training AI models for automated ornament classification and heritage documentation, Supporting ethnographic research on regional design variations and  Applications in virtual try-ons for e-commerce and digital museum archives.

Files

Steps to reproduce

Images were captured using Redmi Note 14 Pro (200 MP), Redmi Note 14 (108 MP), and Redmi Note 12 Pro+ (200 MP). Locations: New Market and Gausia Market, Dhaka, Bangladesh. Applied Pre-processing Techniques: Resolution Rescaled to 640x640 pixels (aspect ratio 1:1). Rotation 90°, -90°, 180° These pre-processing techniques were applied using FastStone Photo Resizer 4.4. Used Augmentation Techniques include: Rotation: Between (±15°) Flip: Horizontal flip and Vertical (50% probability). Shear: ±10 °Horizontal and ±10° Vertical. Brightness: Between (±25%) Noise: Up to 0.5% of Pixels These augmentation techniques were applied using RoboFlow.

Institutions

  • Southeast University

Categories

Computer Science, Machine Learning, Engineering Research Data Management, Ornamentals, Deep Learning

Licence