CVPR 2026 Compute Reporting Form - Clarification

Thank you to the CVPR 2026 community for your engagement with our Compute Reporting initiative. We appreciate the thoughtful questions we've received and your willingness to participate in this important pilot program. Your contributions will help our field better understand the computational landscape of our research and promote more transparent AI practices.

Timing and Flexibility: We recognize that with the supplementary materials deadline approaching in three weeks, some authors may have limited time to gather detailed computational data. To address this practical concern while maintaining our commitment to transparency, we're providing the following policy clarification.

CRF Submission Policy: Submitting the Compute Reporting Form (CRF) is MANDATORY for all CVPR 2026 submissions and should be submitted during the supplementary materials submission phase. We've provided a prefilled example to guide you through the process.

Required and Optional Sections:

• Section 1 (Hardware Specifications): Mandatory. This section requests basic information about your hardware infrastructure and takes approximately 5 minutes to complete.
• Section 2 (Task and Compute Reporting): Optional, highly encouraged. This section requests compute costs, performance metrics, and efficiency calculations for your reported results.
• Sections 3 through 4 (Additional Computational Details, W&B Logs, etc.): Optional. These sections provide opportunities to share more comprehensive development information if available.
• Section 5 (Verification and Submission): Mandatory. This section confirms and verifies your submission.

Recognition Awards: Papers that complete Sections 1-3 are eligible for recognition awards (such as the "Efficient CVPR" Badge and "CVPR Compute Gold Star"), while completing all sections (including Section 4 with W&B logs) is required for eligibility for all awards related to compute reporting, including the "CVPR Compute Transparency Champion" award.

Opt-Out Option: If you cannot or prefer not to report computational data—whether due to limited time to gather records, proprietary constraints, competitive considerations, or institutional policies—you SHOULD submit the CRF with an explanation of your situation in the "Additional Comments" section (5.2). This allows you to fulfill the submission requirement while helping us understand barriers to transparency, which will inform future iterations of this initiative.

Completing the Form: For authors who are able to share data, we expect the CRF to be completed with reasonable care and due diligence. While we understand that perfect precision isn't always possible and estimates are acceptable, please provide your best approximation based on available information (cloud bills, cluster logs, training records, etc.). The goal is to capture a reasonable computational profile that benefits the entire community.

Important Distinction - No Impact on Peer Review Process: The requirement to submit the CRF is a procedural requirement for all papers. However, the computational data you report (e.g., GPU hours, FLOPs) will NOT be used in peer review or affect acceptance decisions. Your paper is evaluated solely on scientific merit, regardless of the computational resources reported or whether you choose to complete optional sections. The data you report will ONLY be visible to Program Chairs, NOT to Reviewers or Area Chairs.

Example Use Cases: The goal of this initiative is to understand the disparities that may exist in available/expended computation. In particular, PCs may report correlations between values provided in different CPU/GPU compute fields (e.g., number of cores available, memory available, cloud vs non-cloud resources) and reviewer scores, paper decisions, or potentially the assessment of different aspects of the paper from raw review text. If strong relationships between compute and acceptance are found, this evidence could inform initiatives (e.g., government programs) to make more compute available to under-resourced institutions, or efforts to reduce computational demands (e.g., through more efficient algorithms) and, in turn, reduce energy and climate impact.

For complete details, including answers to common questions, please see the updated Compute Reporting Form - Author Guidelines and FAQ.