1 of 36

HOW TO BE A GOOD REVIEWER?

REVIEWER TUTORIAL FOR ECCV 2024

Aleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, Gül Varol

ECCV 2024 Program Chairs

[Slides adapted from ICCV 2021, CVPR 2021, CVPR 2022, CVPR 2023, CVPR 2024]

2 of 36

THANK YOU FOR SERVING AS REVIEWER!

We are all counting on you:

  • AREA CHAIRS for clearly justified guidance for paper accept/reject decision
  • AUTHORS for fair consideration and constructive feedback
  • COMMUNITY for ensuring that every conference paper teaches something worthwhile

3 of 36

TUTORIAL GOAL: TO UNDERSTAND

The nature of the review process

The role of reviewers in the review process

The expectation on reviewers

What to include/avoid when writing

reviews

1

2

3

4

4 of 36

THE DECISION PROCESS: OVERVIEW

Desk reject;

Recommend reviewers for papers

Confirm all decisions

PAPER SUBMISSION

PAPER DECISION

PROGRAM CHAIRS

AREA CHAIRS

REVIEWERS

AUTHORS

Authors submit

Assign papers to ACs

Provide thorough expert

reviews

Confirm decisions of other ACs within the triplet

Submit rebuttal

Decision to authors

Engage in rebuttal discussion;

Update reviews

Oversee reviewers;

Discuss within AC triplets;

Write consolidation report ("meta-review");

Recommend a decision

1

2

3

4

5

6

7

8

9

10

5 of 36

THE DECISION PROCESS: DETAILS

  • PROGRAM CHAIRS (PCs) assign papers to AREA CHAIRS (ACs).
  • ACs suggest multiple REVIEWERS per paper, with help of TPMS & OpenReview matching.
  • Papers are assigned to REVIEWERS using an optimization algorithm that takes into account AC suggestions, paper load and conflict constraints (between ACs, reviewers, and authors).
  • REVIEWERS submit initial review. ACs check quality of reviews, chase late reviewers, and assign emergency reviewers as necessary.
  • Authors receive reviews and submit rebuttals.
  • Discussion among REVIEWERS and ACs, based on all reviews, rebuttal, and paper. ACs stimulate and moderate the discussion. Reviewers update their ratings and justification.
  • ACs make decisions and write meta-reviews. The decision and meta-review are recorded by the primary AC for each paper and checked/approved by the secondary AC. Primary and secondary ACs discuss borderline papers and difficult cases to reach a decision on every paper. Additional opinions may be sought from other expert ACs after checking for conflicts. In addition to accept/reject decisions, AC triplets provide a roughly ranked list of oral/award nominations to the PCs.
  • PCs verify and confirm ACs decisions.

6 of 36

LEVEL OF DECISIONS

Your job as a reviewer is to provide well-reasoned recommendations to ACs to enable them to make final decisions on all papers

Major advances that will heavily impact the field; will be used by many people, create new capabilities. E.g.: ResNet (CVPR 2016 Best Paper), Mask R-CNN (ICCV 2017 Best Paper), NeRF (ECCV 2020 Best Paper Runner-Up), ...

Outstanding paper: top ~10% of all accepted papers

Unlikely to be significant, or papers with major flaws

Poster (default for all accepted papers): works that add bricks to the cathedral of knowledge; papers introducing useful tools; papers of interest to a subcommunity; creative ideas that are hard to judge but could be promising

AWARD

CANDIDATE

ORAL

POSTER

REJECT

7 of 36

WHAT PAPER SHOULD BE ACCEPTED?

Any paper that, in accordance with ECCV community standards,

  • presents sufficient knowledge advancement that is well grounded,
  • is of sufficient interest to some ECCV audiences who could benefit from it.

Note: ECCV is very inclusive. Historically, rejection solely for out-of-scope is rather rare.

8 of 36

WHY NOT ACCEPT EVERYTHING?

Papers can have a negative impact:

  • Wrong or fraudulent results mislead the field and damage the reputation of the conference
  • Misleading evaluation makes it hard to compare with, kills follow-up
  • Creates bad precedent (weak paper X got in, so this one should too)
  • Fatigue/overload of too many papers, wastes everyone's time

9 of 36

REVIEW FORM OVERVIEW

Explain the key ideas, contributions, and their significance. This is your abstract of the paper. The summary helps the AC and the authors understand the rest of your review and be confident that you understand the paper. You must NOT use an LLM to write the summary.

What about the paper provided value? E.g., interesting ideas that are experimentally validated, an insightful organization of related work, new tools, impressive results, something else? Most importantly, what can someone interested in the topic learn from the paper? Short bullet lists are not enough.

What detracts from the contributions? Does the paper lack controlled experiments to validate the contributions? Are there misleading claims or technical errors? Is it possible to understand (and ideally reproduce) the method and experimental setups by reading the paper? Short bullet lists are not enough.

Carefully explain why the paper should be accepted or not. This section should make clear which of the strengths and weaknesses you consider most significant.

Minor suggestions, questions, corrections, etc. that can help the authors improve the paper. Not crucial for the overall recommendation.

SUMMARY

STRENGHTS

WEAKNESSES

RATING AND JUSTIFICATION

ADDITIONAL

COMMENTS

10 of 36

11 of 36

ECCV 2024 REVIEW FORM DETAILS

1. By taking this review assignment and checking on "I agree" below, I acknowledge that I have read and understood the reviewer guidelines. * (visible to meta-reviewers)

[x] I agree

2. Summary. In 5-7 sentences, describe the key ideas, experimental or theoretical results, and their significance. * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

In your own words, explain the key ideas, contributions, and their significance. This is your version of the abstract of the paper. The summary helps the AC understand the rest of your review and be confident that you understand the paper. In particular, try your best to answer the following questions: (1) What problem is addressed in the paper? (2) Is it a new problem? If so, why does it matter? If not, why does it still matter? (3) What is the key to the solution? What is the main contribution? (4) Did the experiments or the theoretical analysis sufficiently support the claims?

3. Strengths. Consider the significance of key ideas, experimental or theoretical validation, writing quality, data contribution. Explain clearly why these aspects of the paper are valuable. Short bullet lists do NOT suffice. * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

What are the key contributions and why do they matter? Please consider the following aspects: interesting ideas that are experimentally or theoretically validated, a clearly formalized scientific hypothesis, an insightful organization of related work, new tools, impressive results. Most importantly, what can someone interested in the topic learn from the paper, i.e., clearly articulated knowledge advancement? Some specific examples include, but are not limited to - Clear explanations and illustrations - Contributions clearly stated and validated - Innovative problem formulation or solution - New technical insights - Thorough experiments - Thorough theoretical validation (if applicable) - High practical impact - High impact on the research community

4. Weaknesses. Consider the significance of key ideas, experimental or theoretical validation, writing quality, data contribution. Clearly explain why these are weak aspects of the paper, e.g., why a specific prior work has already demonstrated the key contributions, or why the experiments are insufficient to validate the claims, etc. Short bullet lists do NOT suffice. Be specific! * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

What are the aspects of the paper that most need improvement? E.g., - Key ideas and techniques of the paper are difficult to understand - Typos and grammar problems make reading difficult - Contributions are not clearly and accurately stated - Potential importance of contributions is not convincingly described or demonstrated - Paper contains technical or experimental errors - Experiments or theoretical grounding are insufficient to validate the contributions

12 of 36

ECCV 2024 REVIEW FORM DETAILS

5. Paper rating (pre-rebuttal). * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

  • Strong Accept: The paper makes a significant contribution that is validated by experiments or theoretical analysis. You might want to present this paper in a reading group or recommend it to colleagues.
  • Weak Accept: The paper makes contributions that are validated by experiments or theoretical analysis, and you would recommend that an interested person read it. There may be weaknesses, such as lack of highly compelling ideas, poor writing, or gaps in experimentation or theory that do not invalidate the contributions but prevent a more enthusiastic recommendation.
  • Borderline: You cannot decide whether the paper should be accepted. Please clarify in your justification whether the paper is borderline because: you see strong reasons to both accept and reject; or you can’t find any strong reasons to do either; or you are not familiar or engaged enough with the topic to be confident about a recommendation.
  • Weak Reject: The paper has weaknesses such as poor writing, insufficient experimental or theoretical validation, technical flaws, or lack of significant contribution that would prevent you from recommending it to an interested party. However, there are also valuable aspects to the paper, such that you could imagine someone else recommending it, or the paper being brought to acceptable standards through a minor revision.
  • Strong Reject: There are important problems with the paper that invalidate its contributions or make them inaccessible, such as technical errors, claimed contributions not validated by experiments or theory, no new contribution beyond prior work, or major writing problems that prevent understanding the paper.

6. Recommendation confidence. * (visible to other reviewers, visible to meta-reviewers)

  • Very Confident: I am an expert in the paper topic and have carefully considered the paper.
  • Somewhat Confident: I am familiar with the topic and have made an effort to consider the paper, but I might have missed something important.
  • Not Confident: I am limited by expertise and/or time and would not give my recommendation too much weight.

7. Justification of rating. What are the most important factors in your rating? * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

Do NOT fill with “See weaknesses above”. Carefully explain why the paper should be accepted or not. This section should make clear which of the strengths and weaknesses you consider most significant. Explain how you weigh the strengths and weaknesses you identified above. What are the most important concerns that need to be addressed in a potential rebuttal? If your rating is borderline, carefully explain what additional information you are looking for (either from the authors in the rebuttal or from other reviewers during the discussion) that will enable you to make a more concrete recommendation.

13 of 36

ECCV 2024 REVIEW FORM DETAILS

8. Are there any serious ethical/privacy/transparency/fairness concerns? If yes, please also discuss below in Question 9. * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

If you check this flag, please provide details in the following question. Feel free to alert your AC as soon as possible so that the paper can be handled appropriately.

  • Yes
  • No

9. Limitations and Societal Impact. Have the authors adequately addressed the limitations and potential negative societal impact of their work? Discuss any serious ethical/privacy/transparency/fairness concerns here. Also discuss if there are important limitations that are not apparent from the paper. (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

If limitations and social impact are not discussed appropriately, please include constructive suggestions for improvement. Are there any serious ethical, privacy, transparency, fairness concerns such as lack of appropriate licenses, inappropriately sourced data, etc.? If yes, explain.

10. Is the contribution of a new dataset a main claim for this paper? Have the authors indicated so in the submission form? * (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

If the paper is claiming a dataset release as one of its scientific contributions, it is expected that the dataset will be made publicly available no later than the camera ready deadline. The authors should have also appropriately flagged this in the submission form. You can check the submission form answers by clicking on the paper ID.

  • Dataset contribution claim in the paper. Indicated in the submission form
  • Dataset contribution claim in the paper. Not indicated in the submission form
  • No dataset contribution claim

11. Additional comments to author(s). Include any comments that may be useful for revision but should not be considered in the paper decision. (visible to authors during feedback, visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

12. Confidential comments to AC, such as concerns about plagiarism, other ethical violations, or your ability to evaluate the paper (only visible to area chairs and program chairs). (visible to meta-reviewers)

13. If another person wrote or helped you with the review, please identify that person here (only visible to area chairs and program chairs). (visible to meta-reviewers)

If you ask someone to help you with your review, you must check the review for quality and assume full responsibility for it.

14 of 36

ECCV 2024 REVIEW FORM DETAILS

14. Final rating based on ALL the reviews, rebuttal, and discussion (post-rebuttal). * (visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

  • 1. Strong Accept (may be among the top 25% of accepted papers - easy decision to accept)
  • 2. Accept (suitable for acceptance to ECCV)
  • 3. Borderline Accept (leaning towards accept, but would not argue with a reject decision)
  • 4. Borderline Reject (leaning towards reject, but would not argue with an accept decision)
  • 5. Reject (not ready for ECCV)

15. Final justification (post-rebuttal). * (visible to authors after notification, visible to other reviewers, visible to meta-reviewers)

What are the main reasons to accept or reject the paper? Any other considerations? Suggestions to improve in revision or resubmission can also be listed here.

  • Final Reviews Questions

15 of 36

THE NATURE OF THE REVIEW PROCESS

  • The review process will never be perfect
    • It is inevitable that some authors will be dissatisfied with the decisions made
  • We strive for a fair and transparent process for every paper
    • The authors should be clearly informed on why their paper is accepted/rejected
    • Authors may appeal only if they identified potential biases in the process
  • We strive for the most informed decision for each paper
    • Best possible decisions given the paper, reviews, rebuttals, and discussions
  • We strive for a consistent process
    • Follow uniform policies that are clearly announced to the community

16 of 36

THE ROLE OF REVIEWERS IN THE PROCESS

Provide an independent, objective, critical, and comprehensive review

Key: What is the knowledge advancement in the paper?

Discuss with AC and reviewer buddies to (hopefully) reach consensus

Explain clearly the basis of your review and recommendation

It is OK if the reviewers disagree with one another even after discussions

AC will form recommendations weighing in reviews, rebuttals, and discussions

Make your final recommendations with solid justifications

Read the rebuttal and discussions. Do they change your position? Why?

This facilitates the ACs to make final recommendations for the paper

REVIEW

DISCUSS

RECOMMEND

17 of 36

THE EXPECTATIONS ON REVIEWERS

Be constructive to the authors

    • It is necessary to be critical, but avoid offending the authors
    • Instead, suggest how they could make the paper better

Be friendly to your buddy reviewers and ACs

    • People could take diverse views on the same paper
    • Agree to disagree – the discussions do not force consensus
    • Focus the discussions on the technical side and do not take it personally

Be on time, responsible, and responsive

    • Any delay will impose additional workload to your colleagues

18 of 36

YOUR ACs ARE THERE TO HELP YOU

  • If you need any help in the process, your ACs are there to help you
    • Suspicious ethical/mischief concerns should be raised to ACs and PCs
    • Avoid policing the paper directly on such issues
  • ACs know your names
    • They will recognize and help build your reputation if you do good reviews
    • They will not have a good impression of you if you submit sloppy or late reviews
  • ACs nominate reviewers for the Outstanding Reviewers Award
    • Outstanding reviewers are more likely to get invited to serve as ACs in the future.

19 of 36

GUIDELINES

Take the time to do a good review

  • Many experienced reviewers take 2-4 hours per paper. If you're fairly new to reviewing (e.g., graduate student), plan on least 4 hours per paper and take the time to read the paper twice, consider related work, look up unfamiliar techniques, etc.

Be impartial

  • Judge each paper on its own merits. There is no global quota on the number of papers the conference can accept, and no requirement that the acceptance rate in your pile should match the acceptance rate of the conference.
  • Be aware of your own bias. We all tend to assign more value to papers that are relevant to our own research. Try to ignore “interestingness of topic” or “fit to the conference” and focus on whether the paper can teach something new to an interested reader.
  • Try to discount the identity of the authors if you happen to know it (e.g., through arXiv). If you do not already know who the authors are, do not attempt to discover them by searching arXiv.

20 of 36

GUIDELINES

Be specific and detailed

  • Your comments will be much more helpful to the ACs and the authors than your scores.
  • Do not simply give summary judgments (“not novel”, “unclear”, “incorrect”) — justify them in detail!
  • This is particularly important for prior work. It is not OK to simply say “this has been done before – you need to give specific references!

Be professional and courteous

  • Belittling, sarcastic, or overly harsh remarks have no place in the reviewing process.
  • Avoid referring to the authors in the second person (“you”). Instead, use the third person (“the authors” or “the paper”). Referring to the authors as “you” can be perceived as being confrontational, even though you may not mean it this way.
  • Do not give away your identity by asking the authors to cite several of your own papers.
  • Proofread and spellcheck your reviews.

21 of 36

GUIDELINES

Be aware that different kinds of papers require different levels of evaluation

  • Potentially transformative idea: basic proof-of-concept.
  • Established problem, plausible idea: benchmark results.
  • Weird, overly complex, implausible, seemingly incremental: extraordinary results (which need to be scrutinized carefully).
  • Position piece or theory paper: no experiments.

Do your work on time!

22 of 36

ETHICS

Avoid conflicts of interest

  • Contact the PCs at eccv2024-programchairs@ecva.net if you suspect you may be conflicted with one of the authors.
  • Refer to ECCV Submission Guidelines for detailed definition of conflicts.

Protect the authors' ideas

  • Do not show submissions to anyone else, including colleagues or students, unless you have asked them to write a review, or to help with your review.
  • Do not use ideas from submissions you review to develop your own ideas. After the review process, destroy all copies of papers and supplementary material and erase any code you may have written to evaluate the ideas in the papers.

23 of 36

TAKE AWAY POINTS

Respect authors and protect their ideas.

Take the time to do a good review.

Clearly justify your ratings.

Do your work on time!

1

2

3

5

Be constructive.

4

24 of 36

WHAT SHOULD BE INCLUDED IN THE REVIEW?

What problem is addressed in the paper? Is it a new problem? If so, why does it matter? If not, why does it still matter? What is the key to the solution? What is the main contribution? Do the experiments sufficiently support the claims?

What are the key contributions and why do they matter?

What aspects of the paper most need improvement?

Are the assumptions and theories (mathematically) sound? Are the experiments scientifically sound and valid? Is the problem addressed trivial? Did the paper miss important prior work? Has it been done before? If yes, where?

A CONCISE SUMMARY OF THE PAPER

A CLEAR STATEMENT OF STRENGTHS AND WEAKNESSES

A COMPREHENSIVE CHECK OF POTENTIAL FUNDAMENTAL FLAWS IN THE PAPER

25 of 36

WHAT SHOULD BE AVOIDED IN THE REVIEW? COMMON MISTAKES

Arrogance, ignorance, and inaccuracy

Pure opinions

Novelty fallacy

Blank assertions

Intellectual laziness

Policy entrepreneurism

26 of 36

ARROGANCE, IGNORANCE, AND INACCURACY

ARROGANCE

IGNORANCE

INACCURACY

SAFE BEHAVIOUR

SCRIPT:

AUTHORS: We did A by means of B

REVIEWER: The only way to do A is through C (i.e., my way or highway) … …

ERROR: you should know or check

SCRIPT:

AUTHORS: All A are B

REVIEWER: I do not think all A are B

ERROR: you should know or check

SCRIPT:

AUTHORS: A is a ring, not a field

REVIEWER: All rings are field

ERROR: They are NOT ……

Do not provide an opinion on things you do not know about

27 of 36

PURE OPINIONS

SCRIPT:

REVIEWER: This is not good enough for ECCV 2024. Why?

SCRIPT:

REVIEWER: CNN is not that interesting Why?

SCRIPT:

REVIEWER: Adversarial losses guarantees distribution matches. No theoretic proof indeed!

These remarks are pure opinions and not grounded

SAFE BEHAVIOUR

Check if you grounded your statement with a “because … …”

ERROR

28 of 36

NOVELTY FALLACY

SCRIPT:

REVIEWER: This should not be accepted because it is not novel.

Why? By whom and where has this been published before?

Many important things are not that novel. Many novel things are not that important & most really silly things are novel

SAFE BEHAVIOUR

Focus on whether or not the paper presented well-grounded knowledge advancement

ERROR

SCRIPT:

REVIEWER: This should be accepted because it is novel

Why? Provide additional justification.

29 of 36

BLANK ASSERTIONS

SCRIPT: REVIEWER: This has been done before By whom? Where? Why?

Making ungrounded statements. Comment about authors instead of focusing on the paper content

SAFE BEHAVIOUR

Provide evidence to support your assertions, Confine the discussion on the technical content of the paper, not on the authors

ERROR

SCRIPT: REVIEWER: Intrinsic images are not longer important Really? To whom? Why?

SCRIPT: REVIEWER: Experiments on unpublished datasets are not scientific Really? Why?

SCRIPT: REVIEWER: Authors are ignorant/careless/incompetent… Be humble, nobody is perfect.

SCRIPT: REVIEWER: If the authors were smart enough, they would…. Be humble, nobody is perfect.

30 of 36

POLICY ENTREPRENEURISM

SCRIPT: REVIEWER: You must publish your dataset! No such policy!

You imposed your own policies which are 1) not part of the official review policy and 2) against scientific review principles.

SAFE BEHAVIOUR

Make sure you follow common principles in scientific review. Most importantly, focus on whether the paper produced significant knowledge advancement.

ERROR

SCRIPT: REVIEWER: You must beat SOTA! No such policy!

SCRIPT: REVIEWER: You must have a theorem! No such policy!

SCRIPT: REVIEWER: You must beat arXiv papers! No such policy!

31 of 36

INTELLECTUAL LAZINESS

SCRIPT: REVIEWER: Does not beat SOTA so it must be rejected!

Does the paper present sufficient knowledge advancement?

Overemphasize certain factors instead of giving a comprehensive assessment

SAFE BEHAVIOUR

Make sure you follow common principles in scientific review. Most importantly, focus on whether the paper produced significant knowledge advancement

ERROR

SCRIPT: REVIEWER: Beat SOTA so it must be accepted!

Does the paper present sufficient knowledge advancement?

SCRIPT: REVIEWER: Theorem V looks wrong

It is either wrong or correct. You cannot be unsure.

SCRIPT: REVIEWER: There is this error hence it should be rejected

Is the error making the main knowledge advancement invalid?

32 of 36

EXAMPLE OF REVIEWS

  • The following examples are from ICLR, which published reviews in the public domain
  • For ICLR, the review is written as a single narrative, rather than broken into sections as for ECCV/CVPR/ICCV, but the same criteria apply
  • Here we consider the quality of the form, rather than the accuracy of the content, of the review.

33 of 36

REVIEW QUALITY GOOD

Rating: 9: Top 15% of accepted papers, strong accept

Review: First of all this paper was a delight to read. The authors develop an (actually) novel scheme for representing spherical data from the ground up, and test it on three wildly different empirical tasks: Spherical MNIST, 3D-object recognition, and atomization energies from molecular geometries. They achieve near state-of-the-art performance against other special-purpose networks that aren't nearly as general as their new framework. The paper was also exceptionally clear and well written.

The only con (which is more a suggestion than anything)--it would be nice if the authors compared the training time/44 of parameters of their model versus the closest competitors for the latter two empirical examples. This can sometimes be an apples-to-oranges comparison, but it's nice to fully contextualize the comparative advantage of this new scheme over others. That is, does it perform as well and train just as fast? Does it need fewer parameters? etc.

I strongly endorse acceptance.

+ Clearly explains why the paper should be accepted

- Does not contain many details about the contribution or why it is novel, so relies on the AC trusting the reviewer's judgment on these points

Note: though the proposed method does not achieve the best results (according to the review), the paper is highly valued for proposing a more general framework.

+ Indicates that the reviewer tried to think of weaknesses but could not come up with anything that should negatively impact the paper rating

+ Constructive feedback for the authors

Though missing a summary of contribution, the review clearly explains why the paper should be accepted

34 of 36

REVIEW QUALITY Ok BUT NOT GREAT

RATING Rating: 8: Top 50% of accepted papers, clear accept

The paper proposes a framework for constructing spherical convolutional networks (ConvNets) based on a novel synthesis of several existing concepts. The goal is to detect patterns in spherical signals irrespective of how they are rotated on the sphere. The key is to make the convolutional architecture rotation equivariant.

PROS: + novel/original proposal justified both theoretically and empirically; + well written, easy to follow; + limited evaluation on a classification and regression task is suggestive of the proposed approach's potential; + efficient implementation

CONS: - related work, in particular the first paragraph, should compare and contrast with the closest extant work rather than merely list them;

- evaluation is limited; granted this is the nature of the target domain

Presentation:

* While the paper is generally written well, the paper appears to conflate the definition of the convolutional and correlation operators? This point should be clarified in a revised manuscript.

* In Section 5 (Experiments), there are several references to S‘2CNN. This naming of the proposed approach should be made clear earlier in the manuscript. As an aside, this appears a little confusing since convolution is performed first on S42 and then SO(3).

Evaluation:

* What are the timings of the forward/backward pass and space considerations for the Spherical ConvNets presented in the evaluation section? Please provide specific numbers for the various tasks presented.

* How many layers (parameters) are used in the baselines in Table 2? If indeed there are much less parameters used in the proposed approach, this would strengthen the argument for the approach.

Minor Points:

- some references are missing their source, e.g., Maslen 1998 and Kostolec, Rockmore, 2007, and Ravanbakhsh, et al. 2016.

- Figure 5, caption: "The red dot correspond to" --> "The red dot corresponds to"

+ Highlights key ideas and contributions.

- The summary should also include one sentence on experimental setup and one sentence on significance of the contribution

+ Itemizes strengths and weaknesses

- Does not provide enough detail. E.g., what is original about the paper? How is the evaluation limited?

+ Includes clarifications questions and constructive feedback for authors

+ Makes it clear that "Minor Points" are not an important factor in decision

Makes general factors in decision clear and provides detailed feedback to authors, but does not provide adequate explanation for strengths and weaknesses

35 of 36

REVIEW QUALITY BAD

Rating: Rating: 4: Ok but not good enough - rejection

Review:

1. The idea of multi-level binarization is not new. The author may have a check at Section "Multiple binarizations" in [a] and Section 3.1 in [b]. The author should also have a discussion on these works.

2. For the second contribution, the authors claim "Temperature Adjustment" significantly improves the convergence speed. This argument is not well supported by the experiments. | prefer to see two plots: one for Binarynet and one for the proposed method. In these plot, testing accuracy v.s. the number of epoch (or time) should be shown. The total number of epochs in Table 2 does not tell anything.

3. Confusing in Table 2. In ResBinNet, why 1-, 2- and 3- level have the same size? Should more bits required by using higher level?

4. While the performance of the 1-bit system is not good, we can get very good results with 2 bits [a, c]. So, please also include [c] in the experimental comparison.

5. The proposed method can be trained end-to-end. However, a comparison with [b], which is a post-processing method, is still needed (see Question 1).

6. Could the authors also validate their proposed method on ImageNet? It is better to include GoogleNet and ResNet as well.

7. Could the authors make tables and figures in the experiment section large? It is hard to read in current size.

Reference

[a] How to Train a Compact Binary Neural Network with High Accuracy. AAAI2017

[b] Network Sketching: Exploiting Binary Structure in Deep CNNs. CVPR 2017

[c] Trained Ternary Quantization. ICLR 2017

- Cites papers that make the idea “not new”, but it does not say how these methods relate

- Not clear. Because it is not tested by experiments, or that the convergence speed is not different?

- Points 3-6 may help authors improve the paper, but it is not clear if they are a significant factor in the rating to reject

Big problems:

- AC can't make good use of the review without reading the paper, due to lack of justification.

- No strengths listed, which may indicate that reviewer is just looking for reasons to reject.

- Author and AC don't know which of the listed points are important for reject rating.

The review lists only weaknesses and requests for clarification, omitting a summary and justification for decision. Thus, it is unclear to author or AC which of these points are the primary basis for the rating.

36 of 36

ADDITIONAL RESOURCES

https://sites.google.com/view/reviewing-the-review-process/

  • CVPR 2021 Training Referee Movie

http://luthuli.cs.uiuc.edu/~daf/CVPR21Training.html

  • CMT Tutorial with Screenshots to the Interface

https://cmt3.research.microsoft.com/docs/help/reviewer/reviewing-guide.html