Skip to yearly menu bar Skip to main content


CVPR 2025 Changes

Due to the dramatic increase in the number of submissions and the deteriorating review quality, we will make several major changes compared to the previous CVPR editions. Please read these changes carefully:

 

1. If they do not serve in another capacity for the organization of CVPR 2025, all authors are obliged to act as reviewers. The Program Chairs (PCs) will ensure that only qualified authors have papers assigned, and the number of reviews will depend on the level of seniority.

There were about 9k reviewers but more than 30k authors at CVPR 2024. Given the personal benefit authors receive from having a paper accepted at CVPR, we believe that it is unfair to the community when authors do not contribute to the reviewing process. Authors who serve in another capacity for the organization of CVPR 2025 or as General/Program Chairs for ICCV 2025 are exempt from this rule.

ADDENDUM: Some authors of multidisciplinary papers may not be from the field of computer vision and pattern recognition. These authors will not have to commit to reviewing papers. Other exceptions include parental leave and military service. The number of reviews assigned to each author will be commensurate with that author’s level of contribution to CVPR 2025 submissions.

2. If a reviewer is flagged by an Area Chair as “highly irresponsible”, their paper submissions will be desk rejected per the discretion of the PCs

It is expected that reviewers will submit fair and thoughtful reviews on time. An option will be available to ACs to flag a review as “highly irresponsible”, in addition to the “not met expectations” as in the past. The ACs will be directed to use this flag sparingly. Example cases of highly irresponsible behaviors are one-sentence reviews, reviews generated by Large Language Models (LLMs), reviews not relevant to the paper or that miss a substantial portion of the paper. Highly irresponsible flags are NOT supposed to be used in cases where reviewers merely have some misunderstandings, missed small parts of the paper, or hold a different opinion from other reviewers or the area chair. If a reviewer fails to submit their reviews, or a review is flagged as being highly irresponsible, it will enter an oversight process adjudicated by the PCs. The final decision about desk rejection will be decided per the discretion of the PCs.

3. An author cannot submit more than 25 papers.

We expect each author listed on a paper to make a significant and meaningful contribution to the work. A limit of 25 papers should be amply sufficient for every author to submit their scientific contributions. If an author submits more than 25 papers, the Program Chairs (PCs) will desk-reject the papers registered after the twenty-fifth one.

4. Large language models (LLMs) are NOT allowed to be used for writing the reviews nor the meta-reviews at any step. 

You cannot use an LLM to write your review. This is true for any LLM, whether you run it locally or use an API.

This policy includes but is not restricted to:

  1. You can’t ask an LLM to write content for you. The review needs to be based on your own judgment.
  2. You can’t share substantial content from the paper or your review with an LLM. This means, for example, that you cannot use an LLM to translate a review.  
  3. You can use an LLM to do background research or to check short phrases for clarity.

Enforcement:

(Senior) ACs will check (meta-)reviews for LLM policy violations. If a review is flagged as a possible violation, the review will enter the same oversight process as mentioned above. The process will determine not if the review is certainly written by an LLM (which is not yet possible), but if the review is highly irresponsible. If this is found to be the case, the papers submitted by the reviewer will be desk rejected per discretion of the PCs.

5. The reviewers’ names will be visible on OpenReview to the other reviewers of the paper after the final paper decisions have been made.

Waiting for the final decision will avoid authority bias. Names will also be visible to ACs, Senior ACs and PCs, as has been the policy in past years.

6. All the authors must complete their OpenReview profiles. 

We need the complete OpenReview profile for better assignment and conflict detection. Papers whose authors have not completed their profile will be desk rejected.

7. Data Sharing

To improve future review quality, the CVPR 2025 PCs plan to share the CVPR 2025 reviewing data privately with the PCs of other future related venues. This data will include personally identifiable records on each reviewer, including statistics on review assessments and lateness.