IEEE TIP Response to Reviewers: Revision Guide for Methods Papers

The IEEE Signal Processing Society publications FAQ explains that the response-to-reviewers material may be uploaded in Manuscript Files or exposed through the Author's Response control. Follow the fields shown for the specific revision. In the document itself, distinguish reviewer comments from author responses with bold labels or another accessible typography.

Dear Associate Editor,

Thank you for the opportunity to revise our manuscript. We focused on the
three decision-level issues: the distinction from the conference version,
fair comparison with recent restoration methods, and isolation of the proposed
module through ablation and complexity analysis.

Reviewer 1, Comment 1: [Paste the complete comment.]

Response: We agree that the original experiment did not isolate the proposed
operator from the larger backbone. We added a controlled ablation with fixed
training data, parameter budget, schedule, and loss. The results appear in
Table III and Figure 5. The method description is revised on page 6,
lines 181-224, and Algorithm 1 now identifies the operator explicitly.

Quote equations or revised sentences when a small change carries the answer.

Treat the following themes as connected tests of one technical contribution. A fair baseline table cannot rescue unclear novelty, and a clear derivation cannot rescue an evaluation that hides protocol differences.

Novelty and appropriateness

IEEE Signal Processing Society reviewer guidance emphasizes novelty and appropriateness. A response should identify the reusable image-processing contribution and the problem class it serves. Do not define novelty as "first use of model X on dataset Y" when the method itself is unchanged.

If the reviewer sees the work as an application paper, revise the introduction, method schematic, and experiment design so the general method is visible. If the contribution really is application-specific, consider whether the journal claim should be narrower.

Conference-to-journal extension

List the prior conference paper and describe the journal-only contribution precisely: new theory, an expanded algorithm, materially broader experiments, failure analysis, complexity work, or a new problem formulation. A longer literature review and more implementation detail are not enough by themselves.

Use a comparison table that maps each claim, equation, dataset, and experiment to the conference and journal versions. Cite the prior paper in the manuscript and explain the extension in the submission materials.

State-of-the-art baselines

For each requested baseline, state the source, code or reimplementation, training data, preprocessing, evaluation protocol, parameter budget, and metric. Explain deviations from the authors' reported setup.

Do not compare a tuned proposed method with untuned public checkpoints and call the result controlled. When exact reproduction is impossible, say what was available and use the fairest valid comparison.

Ablation and sensitivity

An ablation should isolate the claimed component. Keep data, backbone, optimization, compute, and supervision fixed whenever possible. Report whether the component changes accuracy, perceptual quality, robustness, runtime, or stability.

When modules interact, include a factorial or cumulative design that exposes the interaction rather than removing one component at a time from an already different system.

Complexity and deployment claims

Report parameter count, operations or latency under named hardware and settings, memory where relevant, image size, batch size, and implementation framework. Distinguish training cost from inference cost.

Do not claim real-time performance from one favorable device or low-resolution input without the operating conditions. If deployment was not studied, describe efficiency rather than deployability.

Reproducibility

Update equations, pseudocode, initialization, loss weights, augmentations, stopping rules, random seeds, and data splits. If code is available, ensure the response's command and configuration match the final paper.

For learned methods, report variance or repeated runs when the conclusion could depend on random initialization. For hand-designed algorithms, include convergence conditions and parameter sensitivity where they affect use.

The goal is not to win an argument. It is to make the technical distinction reproducible.

A requested baseline may use inaccessible data, a different task definition, non-public code, or a metric that cannot be reproduced fairly. Use this sequence:

1. State what scientific comparison the reviewer wants.
2. Explain the protocol incompatibility or access limit.
3. Add the nearest fair comparison or a controlled proxy.
4. Disclose the remaining limitation.
5. Remove superiority language that the available evidence cannot support.

Do not use implementation difficulty as a reason to omit a central competitor. A fair response may require additional work or a narrower claim.

Reviewer 2, Comment 4: The improvement may come from the larger backbone rather than the proposed frequency-domain module.

Response: We agree that the original comparison changed both factors. We added a controlled 2 x 2 ablation that crosses backbone size with the proposed module while holding training data, augmentation, loss, schedule, and parameter-matched projection layers constant. The module improves [metrics] in both backbones, while the larger backbone contributes a separate gain. Results appear in Table IV and Figure 6. The experimental protocol is on page 9, lines 286-331, and the revised claim on page 3, lines 72-79 now attributes the gains separately.

This response lets the reviewer see which component earned which claim.

Reconcile:

• equations, symbols, and dimensions across text and pseudocode
• algorithm steps against the released implementation
• dataset versions, splits, exclusions, and preprocessing
• baseline source, training data, and evaluation protocol
• parameter, compute, latency, and memory definitions
• metric direction, units, and statistical summaries
• conference-version and journal-version contribution statements
• tables, figure captions, supplement, code, and response

Generate page and line references from the final clean manuscript. Check that equation and table numbering did not change after layout edits.

Across our IEEE TIP pre-submission and revision reviews, three patterns repeatedly leave novelty or comparison concerns unresolved. These are Manusights observations, not IEEE decision rules.

The IEEE TIP ablation changes several factors at once

The revised table removes the proposed module but also changes the backbone width, training schedule, loss, or parameter count. The result cannot isolate the claimed contribution. We map each IEEE TIP ablation row to data, supervision, architecture, optimization, and compute. A useful response names what is fixed and what changes. When perfect matching is impossible, it reports the residual difference and avoids assigning the full gain to one component.

The IEEE TIP journal extension is a quantity list

Authors may answer the conference-overlap concern with more pages, more datasets, and more citations. That does not show a new technical contribution. We compare the prior and revised IEEE TIP papers at the level of problem definition, equations, algorithm, theoretical result, experiment, and conclusion. The response should identify which journal-only addition changes what a reader can understand, reproduce, or apply. A contribution map is stronger than a paragraph claiming the version is "substantially extended."

The IEEE TIP baseline table hides protocol asymmetry

The proposed method may use extra training data, a larger image size, more test-time processing, or a different metric implementation. In our review work, we trace every headline comparison to dataset version, supervision, checkpoint, preprocessing, and evaluation code. The IEEE TIP response should disclose asymmetry beside the table and, when possible, add a controlled comparison. If the protocols remain incomparable, the manuscript should not use the row to claim state-of-the-art superiority.

We also inspect whether the IEEE TIP response reports only the best run. A small gain can disappear across seeds, datasets, or operating points even when one checkpoint looks favorable. The revised experiment should state the selection rule and show variability when randomness can change the conclusion. If repeated training is impractical, the response should disclose that limit and avoid precision that the evaluation cannot support.

For reconstruction and enhancement work, we compare the numerical table with the visual examples and failure cases. A higher aggregate metric does not settle whether structure is hallucinated, fine detail is lost, or artifacts worsen in a clinically or scientifically important region. The response should connect metric gains to representative outputs and name where the method fails.

These failures make the response longer without making the proposed image-processing contribution easier for the Associate Editor and reviewers to isolate or reproduce.

The response says "added" without reporting the result

State what the new experiment showed and whether it changed the claim. Name the metric, operating condition, and manuscript location instead of asking the reviewer to rediscover the result.

More benchmarks substitute for mechanism

Additional datasets do not isolate a method component. Use ablations and controlled comparisons that hold data, architecture, optimization, and supervision constant wherever the claim depends on one module.

Complexity numbers lack operating conditions

Name hardware, precision, image size, batch, implementation, and measurement method. Separate training cost from inference cost and avoid comparing latency measured under incompatible settings.

Code and manuscript implement different methods

Verify equations, defaults, preprocessing, and evaluation scripts against the final repository state. A reproducibility reply is not complete when the released configuration cannot produce the table cited in the response.

Most major revisions remain conditional, and rejection is still possible when novelty remains a conference-scale increment, baselines are not comparable, or ablations cannot isolate the claimed method. A large experiment appendix does not repair an unclear reusable contribution.

Think twice before resubmitting if the central superiority claim depends on extra data, compute, or supervision that the comparison table does not disclose. Run a controlled experiment or narrow the claim first.

• The Associate Editor's controlling issues are summarized clearly.
• Every reviewer comment is reproduced and answered.
• Each reply states the action, result, interpretation, and exact location.
• Novelty is visible in the equation, algorithm, experiment, and conclusion.
• Conference and journal contributions are mapped precisely.
• Baseline protocols and ablation controls are disclosed.
• Reviewer comments and author replies are visually distinct.
• The clean manuscript, marked copy, response, supplement, and code agree.

Use the free revision readiness scan to check whether the technical evidence answers the revision decision.

We reviewed current IEEE Signal Processing Society author, reviewer, and ScholarOne workflow materials, established rebuttal guidance, and recurring method-comparison problems in Manusights review work. Use this guide when an IEEE TIP revision must make novelty, conference extension, baselines, ablations, complexity, and reproducibility independently auditable.

Reviewed July 12, 2026.

• IEEE Signal Processing Society publications FAQ
• IEEE Signal Processing Society reviewer guidelines
• IEEE Signal Processing Society information for authors
• PLOS Computational Biology: Ten Simple Rules for Writing a Response to Reviewers

Official IEEE sources support the review criteria and workflow. The response maps and failure patterns are Manusights editorial guidance and do not predict an IEEE TIP decision.