qr-dn1.0 / README.md
devmandan's picture
Initial upload: repackaged for HuggingFace with a full datacard
61f0186 verified
|
Raw
History Blame Contribute Delete
2.77 kB
metadata
license: cc-by-4.0
pretty_name: QR-DN1.0  Distorted and Noisy QR Codes
task_categories:
  - image-to-image
  - image-to-text
tags:
  - qr-code
  - denoising
  - restoration
  - barcode
size_categories:
  - 1K<n<10K
configs:
  - config_name: aligned
    data_files:
      - split: train
        path: data/aligned/train*.parquet
      - split: test
        path: data/aligned/test*.parquet
  - config_name: source_qr
    data_files:
      - split: train
        path: data/source_qr/train*.parquet
      - split: test
        path: data/source_qr/test*.parquet

QR-DN1.0 — Distorted and Noisy QRs

A mirror of QR-DN1.0 (Mendeley Data, DOI 10.17632/t2bdr663ms.2), repackaged as parquet with the five aligned image variants joined into single rows.

Released by its authors under CC BY 4.0, which is what makes this mirror straightforwardly permitted with attribution.

Configs

aligned — 1,500 train / 750 test

Five variants of the same sample, one row each. The upstream directories were verified to carry identical stems, set-equal across all five, so the join is exact rather than approximate — no row is dropped or padded.

column upstream directory
target target/
target_quad target quad/
extracted_one extracted One/
extracted_quad extracted Quad/
extracted_voted extracted Voted/

source_qr — 50 train / 25 test

The QR/ directory: source codes named ht1…htN, on a naming scheme unrelated to the aligned samples.

An honest limit on this card

The semantics of the five variants are not documented here, because this is a mirror and the packager did not establish which variant is input, which is ground truth, or what "One", "Quad" and "Voted" denote as methods. The names are preserved exactly as published so they can be matched against the paper. For the authoritative description, go to the DOI. If you want a card that explains the protocol, that should come from someone who has read the paper, not from the directory listing.

Usage

from datasets import load_dataset

ds = load_dataset("devmandan/qr-dn1.0", "aligned", split="train")
row = ds[0]
row["target"], row["extracted_voted"]      # PIL images, same sample

src = load_dataset("devmandan/qr-dn1.0", "source_qr", split="train")

Citation

@misc{qrdn1_2021,
  title     = {QR-DN1.0: A new Distorted and Noisy QRs dataset},
  author    = {Monfared, Milad and Koochari, Abbas and Monshinmotlagh, Radin},
  year      = {2021},
  publisher = {Mendeley Data},
  version   = {V2},
  doi       = {10.17632/t2bdr663ms.2}
}

Licence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/ Please cite the authors above, not this mirror, when using the data.