Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/cvpr-related-work
Use when building the related-work section and positioning for a CVPR paper, covering literature that moves at arXiv speed, concurrent-work etiquette, the CVPR/ICCV/ECCV sibling triangle, double-blind self-citation, verifying that cited "CVPR papers" are actually CVPR papers, and defining deltas against the true nearest neighbors.
Passed 22 of 26 test cases with the skill, and 11 of 26 without it.
and 17 more in the full report
11 fixed11 passed either way4 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
What is the specific definition and target quantity for the 'canonical lineage' shelf in a CVPR related work section?
| Checked | Without skill | With skill |
|---|---|---|
| Output defines canonical lineage as the 2 to 3 foundational ancestor methods of the paper's method family. | Missed | Passed |
| Output identifies the failure mode of omitting canonical lineage as reinventing an older module under a new name. | Missed | Passed |
The instructions your model reads when the skill loads.
At CVPR the related-work problem is not scarcity but velocity: the field's literature updates weekly on arXiv, the venue itself accepted 4,090 papers in the 2026 edition alone, and reviewers are drawn from a 25,000-person pool that includes the authors of whatever you forgot to cite. Position precisely or be positioned by Reviewer 2.
| Shelf | What reviewers check | Failure smell |
|---|---|---|
| Canonical lineage | The 2–3 ancestors of your method family | Reinventing a 2019 module under a new name |
| Current SOTA | The methods atop your benchmark's leaderboard now | Comparing only against year-old baselines |
| Nearest neighbors | The 2–3 papers a hostile reviewer calls "the same" | No explicit delta stated |
| Cross-venue siblings | ICCV/ECCV/NeurIPS/WACV/3DV/SIGGRAPH work on the task | Treating CVPR citations as the whole field |
| Concurrent arXiv | Preprints within a few months of the deadline | Pretending they don't exist |
The nearest-neighbor shelf deserves half the section's words. For each: one sentence of credit, one sentence of mechanism difference, one sentence of consequence ("because [24] rebuilds the volume per frame, it cannot run online; we amortize it"). A delta stated in mechanisms survives review; a delta stated in adjectives does not.
With a November deadline and instant preprinting, collisions are routine. The working etiquette: preprints appearing close to or after the deadline are concurrent — cite them once you know of them, in a short "concurrent work" note, without being obligated to outperform them. Two cautions. First, do not use concurrency as a shield for work that predates your submission by months. Second, during the rebuttal you may be pointed at a "missing comparison" that is actually concurrent — answer with dates, calmly, not with indignation.
CVPR requires citing your own lineage in the third person ("[12] proposed…"), never omitting it — a hole where your prior paper should be identifies you as effectively as a name. Because arXiv posting is legal during review (verified 2026 policy), reviewers may suspect who you are; your job is only to avoid confirming it in the text: no "our previous work [12]", no shared project-page naming, no acknowledgements.
test cases passed, vs 11 / 26 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.