Related Experiment Video
Updated: Jul 21, 2026
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
A real-world framework for automated product recognition and catalog generation: dataset, model, and analysis
Mayank Sah1, Jimson Mathew2, P Dayananda3
1Department of Computer Science and Engineering, Indian Institute of Technology, Patna, India. mayank_2221cs09@iitp.ac.in.
Scientific Reports
|May 12, 2026
Summary
This study introduces a large-scale Indian grocery dataset for computer vision, addressing limitations in existing datasets. The new dataset and an efficient product identification model enable better automated retail management.
Area of Science:
- Computer Vision
- Retail Technology
- Machine Learning
Background:
- Traditional barcode systems for grocery identification are labor-intensive and inefficient.
- Existing computer vision datasets for groceries lack scale, diversity, and real-world variability.
- Automated product recognition using computer vision can leverage existing retail surveillance infrastructure.
Purpose of the Study:
- To introduce a large-scale, diverse grocery dataset from multiple Indian retail stores.
- To present a lightweight product identification pipeline for efficient and accurate automated recognition.
- To provide a benchmark for evaluating computer vision models in realistic retail settings.
Main Methods:
- Collected over 13,000 images from 349 product categories across eight stores in India.
- Developed a lightweight product identification pipeline utilizing omni-scale feature learning.
- Evaluated the model on the proposed dataset, considering challenges like occlusion and viewpoint variation.
Main Results:
- The proposed model achieved a mean Average Precision (mAP@0.50) of 58.3%, precision of 72.9%, and recall of 77.9%.
- The model demonstrates competitive performance with a compact architecture, balancing efficiency and accuracy.
- The dataset captures practical retail challenges, offering a more realistic evaluation environment.
Conclusions:
- The presented dataset serves as a valuable, diverse benchmark for advancing automated grocery recognition.
- The efficient detection framework is suitable for practical deployment in retail environments.
- This work facilitates more robust and scalable computer vision applications in retail management.
