Unitail

The United Retail Datasets and Challenges for Detecting, Reading, and Matching in Retail Scene


News

  • [Jun 2024] The Unitail-Det is not publicly accessible to prevent unauthorized commercial use! For academic purposes, please contact fangyic@andrew.cmu.edu
  • [Nov 2022] The training and testing data for Unitail-OCR text detection and recognition tasks is available easy to use
  • [Nov 2022] We release the code based on MMOCR that supports Unitail-OCR
  • [July 2022] The article of Unitail has been accepted to ECCV2022!
  • [May 2022] The evaluation server for product detection and matching is online.
  • [April 2022] Unitail-Det v1.0 released with all images and annotations for training and validation.
  • [April 2022] Unitail-OCR v1.0 released with all images and annotations for training and validation.
  • [March 2022] The Unitail website is released.

Overview

The United Retail Datasets (Unitail) is a large-scale benchmark of basic visual tasks on products that challenges algorithms for detecting, reading, and matching. It offers the Unitial-Det, with 1.8M quadrilateral-shaped instances annotated; and the Unitial-OCR, containing 1454 product categories, 30k text regions, and 21k transcriptions to enable robust reading on products and motivate enhanced product matching.
All images and their associated annotations in Unitail can be used for academic purposes only.

Citation

    
        @InProceedings{Chen2022unitail,
        author = {Chen, Fangyi and Zhang, Han and Li, Zaiwang and Dou, Jiachen and Mo, Shentong and Chen, Hao and Zhang, Yongxin and Ahmed, Uzair and Zhu, Chenchen and Savvides, Marios},
        title = {Unitail: Detecting, Reading, and Matching in Retail Scene},
        journal = {European Conference on Computer Vision},
        year = {2022}
        }