{"id":"9a576386-bab5-48df-b928-3af322c10ebf","arxiv_id":"2501.18905","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A comparative benchmark of three quantum image encoding methods shows no single best technique, with FRQI using fewer qubits but much deeper circuits and lower finite-shot accuracy.","lead":"This paper benchmarks three methods for loading classical image data into quantum circuits: Qubit Lattice, Phase Encoding, and FRQI. It compares their runtime, circuit size, accuracy, and noise metrics on simulators and one IBM quantum processor.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"FRQI practicality recommendation is not supported by the reported data: at matched input sizes FRQI has the worst precision, and at 16x16 the circuit fails on IBMQ.","rationale":"The reader's verdict is CONDITIONAL, and my concern does not move that verdict: the paper still has reference value as a benchmark of three known encodings, but the central recommendation for FRQI is not derivable from the evidence as presented. The reader's weakest assumption focuses on statistical validity (single image, no shot counts, undefined error at zero-valued pixels). I agree that this is a real issue, and it undermines confidence in the precision values. However, I think the more fundamental problem is that even taking Table I at face value, the numbers contradict the 'practical balance' conclusion: FRQI is the least accurate method at the sizes where all three can be compared, and its only clear advantage is width, while its depth is enormous and eventually prevents execution on real hardware. The paper makes a policy-like recommendation without a stated metric weighting, so the conclusion is an unsupported assertion rather than an inference. A concrete fix is to specify the inversion operation and run a matched-size, repeated-trial comparison; if the numbers confirm FRQI's low precision, the recommendation must be scoped down. This aligns with the reader's conditional acceptance but adds a substantive reason that is independent of sampling noise. I credit the paper for providing runtime and circuit-characteristic trends, and for attempting a QPU comparison; those parts are useful. The unary operation is also undocumented, which is a reproducibility gap that should be closed regardless of the recommendation. Overall, the verdict remains CONDITIONAL: acceptable after removing or sharply qualifying the FRQI suitability claim and adding the missing experimental details.","tokens_in":12480,"tokens_out":4067,"duration_ms":41509,"concrete_test":"Specify the inversion circuit for each encoding (e.g., for FRQI, the gate sequence that maps theta to pi-theta, and for Qubit Lattice/Phase the equivalent rotation inversion) and re-run the correctness benchmark on a fixed 4x4 input using 10,000 shots and at least 20 random images, reporting precision as mean plus/minus standard deviation for all three encodings. Also transpile and attempt to run the 8x8 FRQI circuit on the same IBMQ backend. If FRQI precision remains below Phase and Qubit Lattice at 4x4, and 8x8 still exceeds hardware limits, the 'practical balance' recommendation is not supported and should be removed or explicitly restricted to a narrow 'fewest qubits' criterion.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The actionable central claim is that FRQI offers a practical balance of accuracy and minimal qubit requirements for near-term quantum image processing. This claim requires some trade-off assessment across width, depth, accuracy, and hardware runnability, but no such decision rule is given, and the reported numbers point the other way. At the only matched conventional size, 4x4, FRQI precision is 31.25%, versus 43.75% for Qubit Lattice and 68.75% for Phase Encoding; at 2x2 it is 75% versus 100% for both. The flagship 16x16 FRQI experiment reports only 20.46% precision and mean error 4.34 on a 0-255 scale, and after transpilation the circuit could not run on the IBMQ backend because its depth exceeded hardware capacity. The only metric where FRQI clearly wins is qubit width. Without a stated weighting that says width is overwhelmingly decisive and accuracy can be sacrificed, 'practical balance' is an assertion, not a finding. The conclusion also transitions from 'no encoding universally excels' to a specific recommendation, which is a logical leap. Additionally, the unary inversion operation used to generate the accuracy data is never described, so the accuracy values cannot be independently reproduced or checked. These issues are more load-bearing than the statistical concerns alone, because even if the statistics were perfect, the reported data would not license the FRQI recommendation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper compares three classical-to-quantum data encoding techniques—Qubit Lattice, Phase Encoding, and FRQI—for grayscale image data, measuring encoding runtime, circuit width and depth, reconstruction precision and error, Hellinger fidelity, SupermarQ features, and backend-dependent accuracy and runtime. The authors report that no encoding technique excels across all metrics and recommend FRQI as a practical balance of accuracy and minimal qubit requirements for near-term quantum image processing. The study includes a public code repository and experiments on simulators as well as on IBM Quantum hardware.","tokens_in":12603,"tokens_out":4180,"duration_ms":88275,"significance":"The comparative scope is timely and relevant for practitioners working on hybrid classical-quantum application design. The paper's use of standard tools (Qiskit, SupermarQ metrics, Hellinger fidelity) and its inclusion of real IBMQ results are strengths, and the public code repository supports reproducibility. However, the headline recommendation about FRQI is not supported by the reported data: at matched input sizes FRQI has the lowest precision, and at the largest size the circuit cannot run on the IBMQ backend after transpilation. The statistical basis of the main comparison—single trials, no confidence intervals, and an error metric undefined for zero-valued pixels—is too weak to license a practical recommendation. The broader claim that no encoding universally excels is plausible, but the evidence as presented is not strong enough to distinguish that finding from a qualitative statement about the three methods.","major_comments":[{"comment":"The precision and mean error values in Table I are reported for a single randomized image per input size, with no repeated trials, no confidence intervals, and no stated shot count for the simulation experiments. Because precision values such as Qubit Lattice at 43.75% for 4x4 and 56% for 5x5 differ by amounts that are plausibly sampling noise, the current data do not support quantitative comparisons among encodings. Please report repeated trials with error bars and state the shot count for every table entry.","section":"Table I and Sec. V-C"},{"comment":"The error formula E = |V_observed - V_expected| / V_expected is undefined whenever V_expected = 0. For randomized grayscale images, each pixel has probability 1/256 of being zero, and the paper does not state how such pixels were handled before computing the mean error. This affects the numeric values in Table I and the accuracy curves in Figures 6 and 9; please specify the convention used (for example, excluding such pixels, clamping the denominator, or treating them as a special case).","section":"Sec. IV-B2"},{"comment":"The recommendation that 'FRQI offers a practical balance of accuracy and minimal qubit requirements' is not licensed by the reported data. At the only matched conventional size, 4x4, FRQI precision is 31.25% versus 68.75% for Phase Encoding; at 16x16 the FRQI circuit has 20.46% precision and fails to run on the IBMQ backend after transpilation. With no decision rule or weighting that makes qubit width overwhelmingly more important than accuracy, this statement is an assertion rather than a finding. Please either add a trade-off analysis that quantifies the relative importance of width, depth, accuracy, and hardware runnability, or remove the recommendation.","section":"Conclusion and Table I"},{"comment":"The unary operation used to generate the reconstruction data—'invert the pixel value'—is never defined at the mathematical or circuit level, and the decoding formulas in Eqs. (5) and (11) are not accompanied by the inversion step. Without a precise description of the operation and the measurement-to-value mapping, the accuracy and precision values cannot be independently reproduced or checked. Please provide the explicit gate sequence and the exact inversion formula used.","section":"Sec. IV-C and Table I"},{"comment":"The input-size ranges are disjoint for the three encodings: Qubit Lattice and Phase Encoding are evaluated only up to 5x5, while FRQI is evaluated at 2^n sizes. Consequently, the fidelity comparison in Figure 7 compares different problem sizes, and the only size at which all three methods are directly comparable is 2x2 (with 4x4 available for the precision comparison). Please restrict cross-method comparisons to matched input sizes, or clearly label unmatched comparisons as scaling trends rather than head-to-head differences.","section":"Sec. IV-C and Fig. 7"}],"minor_comments":[{"comment":"The phrase 'with diagram known as a Bloch Sphere' should read 'with a diagram known as a Bloch sphere'.","section":"Sec. II-A"},{"comment":"The normalization factor is written as 1/2^n, but the sum runs over 2^{2n} basis states; the notation is ambiguous because n is not defined as the number of position qubits in this equation. Please clarify the role of n in the normalization.","section":"Sec. III-C, Eq. (8)"},{"comment":"The sentence 'FRQI performs better than the Qubit Lattice and Phase Encoding models as it can process all pixels of the image simultaneously' appears in the description of the method but is not supported by the later precision results. Consider moving this claim to the discussion or removing it.","section":"Sec. III-C"},{"comment":"The parallelism formula is missing a closing parenthesis, and the liveness definition contains the typos 'quibit' and 'otehrwise'; these should be corrected.","section":"Sec. IV-B6"},{"comment":"The probabilistic experiment is described as measuring the correlation between shot count, accuracy, and runtime, but Figure 6 appears to plot only two quantities; please specify the axes and describe how the shaded 'area of interest' is computed from the accuracy/runtime ratio.","section":"Sec. IV-B5 and Fig. 6"},{"comment":"The runtime comparison in Figure 4 uses different problem sizes for FRQI than for Qubit Lattice and Phase Encoding, so the statement that FRQI is fastest conflates encoding method with input size; consider plotting runtime per pixel or per qubit on matched sizes.","section":"Sec. V-A and Fig. 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for cs.ET and has a useful public artifact. My main reservation is that the conclusion overreaches the reported data: the FRQI recommendation should either be removed or supported by a proper multi-objective trade-off analysis. The missing repeated trials, undefined zero-value handling, and unspecified inversion operation are fixable but require additional experiments and a revised manuscript before the central claims can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a straightforward benchmark of three known encoding methods (Qubit Lattice, Phase Encoding, FRQI) on a common workflow, with new measurements of runtime, circuit width/depth, fidelity, SupermarQ features, and one QPU run. That is a reasonable thing to do, and the multi-metric template is useful for practitioners in quantum image processing. The writing is clear, the code is promised, and the authors are honest about FRQI's probabilistic nature and about the 16x16 transpilation failure on IBM hardware. Credit where due: this is a real experimental comparison, not a theory paper, and the raw data have reference value.\n\nBut the central actionable claim — that FRQI offers a practical balance of accuracy and minimal qubit requirements — does not follow from the reported numbers. At the only matched sizes (2x2 and 4x4), FRQI has the worst precision of the three: 75% vs 100% at 2x2, and 31.25% vs 43.75% and 68.75% at 4x4. Its only clear win is qubit width. The paper never states a weighting that would make width decisive when accuracy is substantially worse, so 'practical balance' is an assertion, not a finding. The stress-test note is right about this, and the issue is load-bearing, not cosmetic.\n\nThere are additional methodology gaps. The unary operation used to generate the accuracy data is never described, so the numbers cannot be independently reproduced. The relative error formula E = |V_observed - V_expected| / V_expected is undefined when V_expected = 0, and the paper does not say how those pixels were handled. Table I reports single runs without error bars or stated shot counts, and the input sizes for FRQI (2, 4, 8, 16) do not overlap with Qubit Lattice/Phase (2, 3, 4, 5) beyond 4x4, so the cross-method scaling comparison is incomplete. These are fixable, but they matter because the paper's contribution is empirical.\n\nWho is this for? Someone choosing an encoding for a small quantum image processing experiment will find the tables and figures a useful starting point, and the SupermarQ analysis is a good example of how to think about circuit structure. But the conclusion should be read as a suggestion, not a supported recommendation. The paper deserves a serious referee, and it could be acceptable after revision: add error bars, describe the operation, handle the zero-division case, and either replace the FRQI recommendation with a stated trade-off rule or drop it. As is, I would not cite it, but I would not mind seeing it in a workshop.","headline":"Useful benchmark data, but the FRQI recommendation in the conclusion is an overreach that the paper's own numbers do not support.","tokens_in":13279,"tokens_out":2400,"would_cite":false,"duration_ms":25081,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper measures three quantum image encodings and concludes that none dominates; FRQI best balances accuracy and qubit count for near-term tasks.","keywords":["quantum encoding","quantum image processing","FRQI","Qubit Lattice","Phase Encoding","circuit depth","quantum benchmarking","hybrid classical-quantum computing"],"falsifier":"Run the same benchmarks with repeated random images per size (for example, 20 trials per size) and compute confidence intervals for precision and mean error; if the reported differences between methods at a given size collapse to noise, the ranking claim is not supported. Also inspect how the relative-error formula handled pixels whose expected value was zero; if those pixels were dropped or assigned arbitrary errors, recomputing the mean error with a defined convention would show whether the accuracy comparisons still hold.","tokens_in":12149,"feed_emoji":"⚛️","tokens_out":10673,"duration_ms":94780,"temperature":0.7,"pith_summary":"This paper tries to establish how much classical-to-quantum encoding costs in practice by running three common image-encoding methods through the same workflow: encode a grayscale image, invert its pixels, decode, and compare. It finds the methods are complementary rather than ranked: Qubit Lattice and Phase Encoding use one qubit per pixel and stay accurate at low shot counts, while FRQI uses far fewer qubits but needs deeper circuits and many shots. The conclusion is that no encoding technique wins on every metric, so the right choice depends on the algorithm, the circuit size, the shot budget, and the hardware. This matters because encoding overhead is often left out of quantum-advantage estimates, and the paper quantifies that hidden cost.","feed_headline":"No encoding wins on every metric; FRQI balances qubits and accuracy","feed_subtitle":"Shots, circuit depth, and noise each favor a different encoding; pick by algorithm and hardware.","key_machinery":"The load-bearing machinery is a comparative benchmark built around a single unary workload: encode a grayscale image, invert every pixel value, decode, and measure. Three encodings are implemented as gate circuits: Qubit Lattice applies one Ry rotation per pixel on its own qubit; Phase Encoding applies an H-Rz-H sandwich per pixel; FRQI builds a uniform superposition over pixel positions and then applies a multi-controlled Ry rotation per pixel to encode intensity. The comparison metrics are circuit width and depth, encoding runtime, precision and mean relative error, Hellinger fidelity between ideal and measured distributions, and the circuit-structure features of communication, critical depth, entanglement ratio, parallelism, and liveness. That combination lets the paper attribute accuracy differences to shot count and circuit structure rather than to the specific pixel-inversion operation.","core_discovery":"The paper's central claim is that classical-to-quantum encoding is a multi-dimensional trade-off. Qubit Lattice and Phase Encoding each consume one qubit per pixel, keep circuit depth low, and reconstruct accurately at low shot counts, while FRQI compresses a $2^n \\times 2^n$ image into $2n+1$ qubits at the cost of deep multi-controlled-rotation circuits, lower fidelity at practical shot counts, and precision that improves only as the shot count grows. On the tested real device, the largest FRQI circuit (16×16, depth 2554) could not be executed after transpilation, which the paper takes as evidence that hardware limits currently cap the practical input size. The paper concludes that no encoding method wins on every metric, and that FRQI offers a practical balance of accuracy and minimal qubit requirements for near-term quantum image processing when qubits are scarce.","pith_inferences":["The paper leaves the zero-expected-value cases in its relative-error formula undefined; because random grayscale images contain zero-valued pixels with probability 1/256 per pixel, re-running the error analysis with an explicit convention could shift the reported mean errors and the accuracy ranking.","The single-random-image-per-size design means the paper has not demonstrated stable differences between Qubit Lattice and Phase Encoding at specific sizes; repeated trials with confidence intervals are the natural next experiment to test whether the precision swings (for example, Qubit Lattice 43.75% at 4×4 versus 56% at 5×5) are real.","The same trade-off pattern a reasonable reader would predict for other real-valued data—one-qubit-per-value angle encodings versus logarithmically-few-qubit amplitude encodings—is likely to hold, but the paper only shows it for grayscale images and a single inversion operation, so extrapolation to other data types is an inference, not a result.","FRQI's 'practical balance' claim implicitly assumes that shot count can be raised to recover accuracy without exceeding runtime or decoherence budgets; on current hardware the window in which that balance holds may be narrow, since the largest tested circuit already exceeded device capacity."],"forward_implications":["Practitioners should include encoding and decoding overhead, plus the shot count needed for a target accuracy, when estimating whether a quantum image-processing task will actually run faster or cheaper than a classical pipeline.","Qubit Lattice and Phase Encoding remain reasonable choices for small images when qubits are plentiful and per-shot accuracy matters most.","FRQI's low qubit count makes it the fastest encoder at a given image size, but its accuracy improves only with more shots, which inflates total runtime; the optimal shot budget is set by the accuracy-versus-runtime trade-off.","Circuit-structure features such as critical depth and entanglement ratio point to where noise will hurt: FRQI's many controlled two-qubit gates align with its lower fidelity and its failure to run at 16×16 on the real device.","No single encoding should be treated as a default for all quantum image processing; the paper's data support matching the encoding to the workload and hardware constraints."],"supporting_citations":[{"why":"introduces the Qubit Lattice encoding with one qubit per pixel and is the baseline for the qubit-per-pixel comparison.","marker":"[36]"},{"why":"defines the FRQI representation and its polynomial preparation, the main competing encoding under study.","marker":"[18]"},{"why":"provides the quantum circuit construction and execution environment used for all experiments.","marker":"[28]"},{"why":"supplies the circuit-structure metrics (communication, critical depth, entanglement ratio, parallelism, liveness) used to compare noise characteristics.","marker":"[35]"},{"why":"provides the simulator backends used for state-vector and noisy simulation runs.","marker":"[16]"},{"why":"is the source the FRQI encoder logic is adapted from for the controlled-rotation preparation.","marker":"[15]"},{"why":"provides the real quantum device backend used for the hardware comparison and the observed 16×16 size limit.","marker":"[8]"},{"why":"defines the Hellinger fidelity measure used to compare ideal and measured probability distributions.","marker":"[17]"}],"fun_headline_variants":["No winner in quantum encoding: each method trades qubits, depth, and noise","FRQI saves qubits but needs deep circuits, limiting real quantum use","Classical-to-quantum encoding: a trade-off, not a single best choice","Qubit efficiency vs. circuit depth: the quantum encoding dilemma","Quantum encoding: no free lunch—each method sacrifices something"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison assumes that the precision and error numbers measured from a single random image at each size are stable enough to rank the encoding methods, even though the experiments were not repeated and the relative-error metric is undefined for zero-valued pixels.","fun_headline_variants_meta":{"raw":{"variants":["No winner in quantum encoding: each method trades qubits, depth, and noise","FRQI saves qubits but needs deep circuits, limiting real quantum use","Classical-to-quantum encoding: a trade-off, not a single best choice","Qubit efficiency vs. circuit depth: the quantum encoding dilemma","Quantum encoding: no free lunch—each method sacrifices something"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000609,"raw_usage":{"total_tokens":2782,"prompt_tokens":835,"completion_tokens":1947,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":451,"completion_tokens_details":{"reasoning_tokens":1850}},"tokens_in":451,"tokens_out":1947,"duration_ms":17320,"temperature":1.0,"reasoning_tokens":1850,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T21:59:01.809816+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same benchmarks with repeated random images per size (for example, 20 trials per size) and compute confidence intervals for precision and mean error; if the reported differences between methods at a given size collapse to noise, the ranking claim is not supported. Also inspect how the relative-error formula handled pixels whose expected value was zero; if those pixels were dropped or assigned arbitrary errors, recomputing the mean error with a defined convention would show whether the accuracy comparisons still hold.","supporting_citations":[{"cited_title":"Stor- ing, processing, and retrieving an image using quan- tum mechanics","cited_arxiv_id":null,"evidence_quote":"introduces the Qubit Lattice encoding with one qubit per pixel and is the baseline for the qubit-per-pixel comparison."},{"cited_title":"A flexible representation of quantum images for polyno- mial preparation, image compression, and processing operations","cited_arxiv_id":null,"evidence_quote":"defines the FRQI representation and its polynomial preparation, the main competing encoding under study."},{"cited_title":"SupermarQ: A Scalable Quan- tum Benchmark Suite","cited_arxiv_id":null,"evidence_quote":"supplies the circuit-structure metrics (communication, critical depth, entanglement ratio, parallelism, liveness) used to compare noise characteristics."},{"cited_title":"Qiskit API reference - Qiskit AerSimulator","cited_arxiv_id":null,"evidence_quote":"provides the simulator backends used for state-vector and noisy simulation runs."},{"cited_title":"Flexible representation of quantum images (frqi) and novel enhanced quantum representation (neqr)","cited_arxiv_id":null,"evidence_quote":"is the source the FRQI encoder logic is adapted from for the controlled-rotation preparation."},{"cited_title":"IBM Quantum Compute Resources","cited_arxiv_id":null,"evidence_quote":"provides the real quantum device backend used for the hardware comparison and the observed 16×16 size limit."},{"cited_title":"Qiskit API reference - Qiskit Hellinger Fidelity","cited_arxiv_id":null,"evidence_quote":"defines the Hellinger fidelity measure used to compare ideal and measured probability distributions."}],"review_version":1}