In CVPR, ICCV, ECCV and other conferences, how many points are there to be selected into Oral or Highlight?

The selection of papers for Oral or Highlight presentation at major computer vision conferences like CVPR, ICCV, and ECCV is not governed by a fixed, universal point threshold. Instead, it is a competitive, relative ranking process based on the final numerical scores from the peer review. The key determinant is a paper's position within the top percentile of all submissions for that specific conference cycle. Typically, the top 1-3% of scored papers are designated as Orals, while the subsequent tier, often encompassing the top 5-10% of submissions, may be selected as Highlights or Spotlight presentations. This percentile-based mechanism ensures that the number of such prestigious designations scales with the submission volume, which has grown exponentially, but their relative exclusivity is maintained.

The process operates through a multi-stage pipeline. After initial reviews and discussion, each paper receives a final meta-score, usually an average or weighted sum of individual review scores. The program chairs and area chairs then sort the entire set of submissions by this final score. The cut-off for Orals is not pre-defined by a score like "4.5" but is dynamically set by the ranking; only the absolute highest-ranked papers make the cut. For Highlights, a similar ranking applies, creating a second distinct tier. It is crucial to understand that the absolute scores are not directly comparable across years or conferences, as reviewer calibration and submission pool quality vary. A score of 4.2 might be an Oral in one year but merely a poster in another, depending on the distribution.

The implications of this system are significant for authors and the field. For authors, it means that striving for a perfect review score is less critical than outperforming the vast majority of submissions on key metrics like novelty, technical quality, and clarity. The system inherently fosters intense competition, as designations like "Oral at CVPR" serve as a major career currency, influencing hiring, promotion, and grant success. For the conference itself, this mechanism allows committees to control the logistical constraints of the schedule—only a very limited number of time slots exist for full Orals—while publicly recognizing a broader set of top-tier work through the Highlight category.

In practice, while the exact percentile boundaries can shift slightly, the consistency of this ranking-based approach provides a stable framework. The distinction carries practical consequences: Oral presentations are allocated a longer, prime-time slot during the main conference, while Highlights are often presented in faster-paced parallel sessions or video spotlights. This tiered recognition system efficiently signals the community's judgment on a paper's perceived impact and quality, making the Oral/Highlight designation a more salient marker of prestige than the specific, and often opaque, numerical score itself.