Corpus 1,272 assessed · 1,173 scored · 643 reproduced ≥75 · 168 flagged ·∅ 74.1/100
← New search

Quantum pixel representations and compression for N-dimensional images.

Sci Rep · 2022
L1 87/100 3/4
Why this verdict

The main results reproduced: recomputed values matched the published ones within tolerance.

Reproduced on the brainbox compute brainarbeit.com
✓ What held up
  • Same input data as the authors
  • Reported values were directly comparable
  • No relevant deviation in data/preprocessing
  • No authors-side cause for any deviation
  • Any deviation was negligible
What did not (or only partly)
  • 🟡Reported values were not (fully) derivable from the shared data
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
87/100
Reproducibility score
0.7 SD above mean
vs. all fields · 1173 studies
🎯 Scores higher than 72% of all assessed papers rank 301 of 1173 scored

A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.

Reproduction agent’s raw note

DESCRIBED WELL ENOUGH -> 1:1 reproduction. QPIXL++ (C++/CMake, commit c2a1b1f; core algorithm byte-identical to paper-era v0.1.0/Zenodo) built and run on «our HPC» compute nodes after a one-line portability patch (#include <algorithm> for GCC>=12; no algorithmic effect). The two headline gate-complexity claims reproduce EXACTLY to the gate (MNIST: 1024 Ry/1024 CNOT/11 qubits; ceramic: 65,536/65,536/17 qubits). The Ry-vs-CNOT compression relationship reproduces exactly at every one of 9 compression levels (Ry%==setting, CNOT%<Ry%). The ceramic reconstructions (Fig 6) match the deposited reference PNGs at 42-54 dB PSNR and are visually indistinguishable. The statevector simulation matches the algebraic reconstruction within integer rounding. NOT pixel-verified: the MNIST Fig-5 panels, because the paper/repo deposited only the MNIST input, not its reconstructions (reproduced by running the pipeline + verified lossless@0% + correct degradation instead). No value appears fabricated; every reported number is re-derivable from shipped code+data.

💻 Code ↗ 🗄 Data: 10.5281/zenodo.5557893

These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.

Assessment versions

Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.

  1. v1 current initial assessment Score 50
    assessed: 2026-06-19 ⛓ faf2ce46a632
✎ I am an author of this paper

Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.

Reason for the rerun

We email you a confirmation link first. The rerun is an objective re-measurement — it cannot change the verdict in your favour, only ask us to look again.

Provenance — full disclosure

When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.

Reproduced
2026-06-25
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-19
no human curator yet
Last updated
2026-08-05

Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.

Deep full-text extraction

Model: sonnet
Founding hypothesis

The paper tests whether quantum image representations (FRQI, NEQR, MCRQI, NCQI and their variants) can be unified under a single general framework (QPIXL) and prepared with substantially fewer quantum gates and no ancilla qubits than prior circuit constructions, while also being compressible without significant loss of image quality.

Core claims
  • QPIXL is a uniform framework that overarches (I)FRQI, (I)NEQR, MCRQI, and (I)NCQI representations resource
  • The QPIXL synthesis method reduces gate complexity for all considered quantum pixel representations without introducing ancilla qubits method
  • QPIXL circuits scale linearly (O(N)) in the number of pixels and use only Ry and CNOT gates, making them practical for NISQ devices finding
  • A proposed circuit/image compression algorithm reduces gates needed to prepare an FRQI state by up to 90% without significantly sacrificing image quality finding
  • QPIXL reduces FRQI state preparation gate complexity from the original O(N^2) (and Khan's O(64N log2 N) with ancilla qubits) to O(2N)/O(N) with no ancilla qubits finding
  • QPIXL reduces IFRQI gate complexity from O(pN log2 N) to O(pN) and removes the need for ancilla qubits finding
  • Algorithms are publicly released as QPIXL++, a Quantum Image Pixel Library built on QCLAB++ resource
  • Definition 1 formalizes the 'Square QPIXL' quantum state as an equal superposition over position basis states tensored with a color-encoding state mechanism
Experimental setups
Assay System Perturbation Readout Platform
quantum circuit gate-complexity analysis/synthesis simulated quantum circuits (theoretical, no specific hardware stated) none number/type of elementary gates (Ry, CNOT) required to prepare image states QPIXL++ (built on QCLAB++)
image compression and quality evaluation example scientific images (FRQI-encoded) compression algorithm applied to reduce circuit gates percentage reduction in gates vs. resulting image quality QPIXL++
Key results
  • FRQI gate complexity reduced from O(N^2) to O(2N) with no ancilla qubits
  • Prior FRQI improvement by Khan achieved O(64N log2 N) but required log2(N)-2 extra ancilla qubits
  • IFRQI gate complexity reduced from O(pN log2 N) to O(pN), removing ancilla qubits
  • Compression algorithm reduces gates needed to prepare FRQI states for example scientific images up to 90%
Key statistics
  • fold_change up to 90% gate reduction (compression of FRQI circuit for example scientific images)
  • other O(N^2) (original FRQI gate complexity)
  • other O(64 N log2 N) (Khan's improved FRQI gate complexity, requires log2(N)-2 ancilla qubits)
  • other O(2N) / O(N) (QPIXL FRQI gate complexity, no ancilla qubits)
  • other O(pN log2 N) to O(pN) (IFRQI gate complexity before/after QPIXL)
  • count n+1 qubits (original FRQI qubit count for N=2^n grayscale pixels)

Statistical methods review

Model: sonnet

A neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.

This is a theoretical computer science and quantum computing paper introducing the QPIXL framework for quantum image representation. The evaluation approach is entirely based on analytical complexity analysis (Big-O gate counts) and computational demonstrations of circuit compression on scientific images; no inferential statistical testing is performed. Results are reported as exact gate counts, asymptotic complexity comparisons between methods, and compression percentages (up to 90% gate reduction) measured on example images.

Replicationunclear Sample sizeNumber of pixels N and qubit counts described analytically; no sample size or statistical power discussion GroupsQPIXL-based implementations vs. prior art implementations for each quantum image representation type Pairingna Randomization/blindingna Dispersionnone
Statistical tests used
Test Applied to n Assumptions
Analytical complexity analysis (Big-O gate counting) Comparison of QPIXL vs. prior FRQI, IFRQI, NEQR, INEQR, MCRQI, NCQI, INCQI circuit implementations na
Computational demonstration of compression ratio Circuit and image compression experiments on scientific images (up to 90% gate reduction stated) not stated
Approaches that could also have been used
  • Image quality after compression is described qualitatively (e.g., 'without significantly sacrificing image quality') without a formal quantitative metric
    Could also: Standard image quality metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), or Mean Squared Error (MSE) could also be used to quantify reconstruction fidelity at each compression level — Quantitative image quality metrics would allow readers to reproduce threshold choices, compare across image types, and understand the trade-off curve between gate reduction and fidelity more precisely
  • Compression performance (up to 90% gate reduction) is demonstrated on example scientific images without reporting variability across images or image types
    Could also: Reporting compression ratios across a benchmark set of images with summary statistics (mean, range, or percentiles) could also characterize the method's typical and worst-case behaviour — A single reported maximum may reflect a best-case outcome; distributional reporting across varied images would convey how consistently the compression performs
  • Complexity comparisons between QPIXL and prior methods are presented as asymptotic Big-O bounds
    Could also: Empirical runtime or gate-count benchmarks on quantum circuit simulators across a range of image sizes N could also accompany the asymptotic analysis — Asymptotic complexity describes large-N behaviour; empirical benchmarks at practically relevant sizes would show whether the constant-factor improvements dominate in the regime of current NISQ hardware
  • No confidence or uncertainty quantification is reported for the compression experiments
    Could also: If experiments involve any stochastic elements (e.g., random image selection or threshold choices), bootstrap confidence intervals or sensitivity analyses across parameter settings could also be reported — Uncertainty estimates allow readers to judge the robustness of the reported compression gains to choices made during experimental setup
Software: QPIXL++ · QCLAB++

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-35546151

Paper: Amankwah, Camps, Bethel, Van Beeumen, Perciano (2022). "Quantum pixel representations and compression for N-dimensional images." Sci Rep 12:7712. DOI 10.1038/s41598-022-11024-y. PMCID PMC9095730.

Code: https://github.com/QuantumComputingLab/qpixlpp (QPIXL++, C++/CMake).

  • Latest master commit (cloned for reproduction): c2a1b1f53c0105b2e051840c85afae01e45c3449 (2025-10-04).
  • Paper-era release tag: v0.1.0 = bdbf53f58109e9d3822f13b724ba5db9c8eb4ad8 (2021-10-08), which is also the Zenodo deposit 10.5281/zenodo.5557893.
  • VERIFIED: the core algorithm headers (include/qpixl/frqi/circuit.hpp, frqi/util.hpp, util.hpp, pgm.hpp) are byte-identical between v0.1.0 and master. The only diff in examples/compressedFRQI.cpp is a one-character OpenQASM syntax fix (qelib1.inc -> qelib1.inc";). => building master reproduces the paper-era algorithm faithfully; gate counts are unaffected.

Data: Zenodo 10.5281/zenodo.5557893 is a snapshot ZIP of the v0.1.0 repo (the code, not a separate data product). The paper's actual figure data ships INSIDE the repo at examples/NSR_Journal_Data/ (added after v0.1.0, present in master):

  • mnist{0,30,60,75,90}.png — Fig 5 reconstructions (MNIST "3", compression 0/30/60/75/90 %)
  • Fig5_{0,50,75,90,95,99}.png — ceramic-composite reconstructions (256x256, compression 0/50/75/90/95/99 %)
  • gray.dat, fwht.dat — grayscale vector and its FWHT coefficients (intermediate). Input images: examples/Example4.pgm (28x28, the MNIST digit). The 0%-compression reference is a lossless reconstruction == the original, so the shipped 0% PNGs double as the inputs for the other compression levels.

Pipeline

compressedFRQI <in.pgm> <out> <compression%> [sim]:

  1. read PGM, zero-pad pixel vector to next power of two N=2^n;
  2. convert grayscale -> FRQI rotation angles;
  3. permuted fast Walsh-Hadamard transform; compression sets the smallest-|coeff| compression% of WHT coefficients to zero; redundant CNOTs removed by parity;
  4. emit OpenQASM 2.0 circuit + a header with gate statistics (nQubits, nGates, nCNOT, nRY, nH, compression setting, CNOT% & Ry% reductions);
  5. reconstruct the (compressed) image via inverse permutation + inverse FWHT, write out.pgm; optionally statevector-simulate with QCLAB++ -> out_sim.pgm. maxgates = 2^(nQubits-1) = N. nH = nQubits-1.

IN SCOPE (pipeline-derived, attempted)

  • C1 Gate complexity / Table 1: QPIXL FRQI = O(N): N Ry + N CNOT, 0 ancilla, n+1 qubits. Check uncompressed (0%) counts: MNIST N=1024 -> ~1024 Ry, ~1024 CNOT, 11 qubits; ceramic N=65536 -> nbQubits 17, gate counts ~2^16=65,536 (Fig 6 text).
  • C2 Fig 5 (MNIST) reconstructions: run at 0/30/60/75/90 %, compare out.pgm pixel-for-pixel to mnist{...}.png.
  • C3 Fig 5 text claim: "reduction in Ry gates is in perfect agreement with the compression ratio, but ... a smaller reduction in CNOT gates." Verify Ry% ≈ set% and CNOT% < Ry% across levels.
  • C4 Fig 6 (ceramic) reconstructions: run at 0/50/75/90/95/99 %, compare to Fig5_{...}.png; check uncompressed gate count = 65,536.
  • C5 QCLAB++ simulation: statevector-simulate a small case (sim=1), confirm out_sim.pgm matches the algebraic reconstruction out.pgm.

OUT OF SCOPE (not attempted; stated)

  • Theoretical complexity proofs and the asymptotic comparison to Le et al. O(N^2) / Khan O(N log N): mathematical, not a runnable pipeline (the >95% reduction in Fig 4 vs Le et al. relies on Le et al.'s un-shipped implementation).
  • Any hardware/real-QPU execution (none claimed for this paper's core results).
  • NEQR and the unrelated 2024/2025 example folders in current master.
Figures / tables: TableFig 5Fig 6
C1a
Reported
~1024 Ry & 1024 CNOT, 11 qubits (MNIST N=1024, Table 1/Fig 5)
Reproduced
1024 Ry, 1024 CNOT, 10 H, 11 qubits
exact
C1b
Reported
65,536 CNOT & Ry, 17 qubits (ceramic 256x256, Fig 6)
Reproduced
65536 CNOT, 65536 Ry, 16 H, 17 qubits
exact
C2
Reported
Fig 5 MNIST reconstructions 0/30/60/75/90%, digit recognizable & degrading
Reproduced
reconstructions produced; 0% lossless (PSNR 52.8 dB); monotone degradation; digit '3' recognizable through ~75%
partial
C3
Reported
Ry reduction == compression ratio; CNOT reduction smaller
Reproduced
Ry% == setting exactly at all 9 levels; CNOT% strictly < Ry% everywhere (e.g. ceramic 50%: Ry 50.0 / CNOT 31.8)
exact
C4
Reported
Fig 6 ceramic reconstructions 0/50/75/90/95/99% (Fig5_*.png deposited)
Reproduced
PSNR vs deposited refs 42-54 dB, max diff <=8/255; visually identical
within tolerance
C5
Reported
QCLAB++ statevector simulation recovers the FRQI image
Reproduced
sim vs algebraic recon PSNR 51.6 dB, max diff 1 (rounding)
within tolerance

Assessments & scoring basis

Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.

🤖 AI curator · claude (ai-curator room) · v1.0 L1 87/100

An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.

🟢1. Data identity
🟢2. Endpoint comparability
🟢3. Location of the main deviation
🟢4. Cause of the deviation
🟡5. Derivability / plausibility
🟢6. Severity of the deviation
🟡7. Core claim
🟡8. Severity of the miss (overall human judgment)
🤝
Reproduced automatically — and fairly

Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.

Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.

🚩 Report an error in this record

Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.

Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.

Reproduction footprint

claude-opus-4-8

Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.

93.2 k
tokens (I/O) · 4 M incl. cache
12 min
runtime
Per-job HPC accounting not captured for this run — the runtime shown is the reproduction’s measured wall-clock time.