devmandan/boofcv-qr-images-v2

BoofCV QR Images V2

A mirror of Peter Abeles' QR Code Performance Benchmark image set from BoofCV, repackaged as a HuggingFace dataset with the corner annotations parsed into structured fields.

487 images: 461 detection images across 14 difficulty categories with 1,153 hand-annotated QR corner quads, plus 26 decoding images with expected payload strings.

Categories

category	images		category	images
nominal	65		rotations	44
glare	50		blurred	42
curved	49		damaged	34
close	40		bright_spots	32
brightness	28		pathological	23
monitor	17		noncompliant	16
shadows	14		lots	7

Fields

field	notes
image	the picture
subset	detection (461) or decoding (26)
category	one of the 14 above, or decoding
corners	list of 4-point polygons, one per QR — every detection image has these
n_qr_codes	QR instances in the image (1,153 total across the detection set)
expected_message	the decoded string, for all 26 decoding images
source_path	original path inside the BoofCV archive

One parsing note that matters

The upstream .txt files use two different formats: some have a SETS header with one line per QR (8 floats = 4 corners), and some have no header and one line per point, four consecutive lines to a QR. Reading only the first format yields corners for 175 of 461 images and silently looks like "the rest are unannotated". Both formats are parsed here, and the point-style files were checked to contain a whole number of quads.

Usage

from datasets import load_dataset

ds = load_dataset("devmandan/boofcv-qr-images-v2", split="test")
hard = ds.filter(lambda r: r["category"] in {"damaged", "curved", "glare"})
multi = ds.filter(lambda r: r["n_qr_codes"] > 1)

Provenance and licensing — read before redistributing

Original source: https://boofcv.org/notwiki/regression/fiducial/qrcodes_v2.zip Author: Peter Abeles (BoofCV).

BoofCV does not state an explicit licence for this image set. The project publishes it openly as a benchmark and describes its images and evaluation code as freely available, which is why it is mirrored here — but "freely available" is not a licence grant, and no licence is or can be granted by this mirror. If you need defined redistribution terms, get them from the BoofCV project.

The parsed annotations and repackaging in this repository are offered under CC BY 4.0. The images are Peter Abeles' work, not the mirror publisher's.

If you are the rights holder and want this mirror removed, open a discussion here and it will be taken down.

Citation

@misc{boofcv_qr_v2,
  title        = {BoofCV QR Code Performance Benchmark, Images V2},
  author       = {Abeles, Peter},
  howpublished = {\url{https://boofcv.org/}},
  note         = {QR detection and decoding benchmark image set}
}

Derived ground truth (qr-decode-bench)

derived_gt/derived_gt.jsonl was ADDED to this repository on 2026-09-01; no existing file was modified (this section was appended to the card). It records, for each image whose payload this dataset does not publish and on which at least two independent QR decoders agreed, what every decoder of a panel decoded on that image — as sha256 digests of the payload bytes, with the payload text beside them only where no identifier detector fires on it — so that a benchmark can score a decoder against what the OTHER decoders established (per-image leave-one-out, quorum 2), never against its own output.

This is truth DERIVED by decoder consensus, not published by the dataset's authors; the benchmark that uses it marks it with †. Agreement between independent implementations is strong evidence, not proof of what a symbol encodes.

• derived at revision 5aa8eebbf1dae45cd5e1ce5d884753b2e5cecbb3 from the data files' recorded inventory (sha256 23d4a7bca8294f2133c53672c1f56676f158e1d871382581ca1403778901da64); the data files are unchanged by this commit
• panel: apple-vision (apple-vision macOS 14.3.1 VNDetectBarcodesRequest rev4), boofcv (boofcv 0.43.1 (PyBoof jar; java 22.0.1)), opencv (opencv 4.12.0 (QRCodeDetector)), opencv-aruco (opencv 4.12.0 (QRCodeDetectorAruco)), quirc (quirc 1.0 (vendored 927d680904dc)), wechat (wechat_qrcode (opencv-contrib 4.12.0; models 8ae41acb,cc49b8c9,e5d36889,e8acfc39)), zxing-cpp (zxing-cpp 3.1.1)
• images with derived truth: 384 of 487 examined (77 without a quorum, 0 voided)
• payloads: 64 established, 63 carried as text, 1 as digest only because an identifier detector fired, 0 as digest only because the payload is the image's own id (image records are keyed by plain of the id)
• format qrb-derived-gt-peer/1: JSON lines — one header (rule, quorum, panel versions, yield, corroboration, privacy scan), one line per payload, one line per image