Agent skill · Documentation

iccv-reproducibility

Use when hardening the reproducibility story of an ICCV paper, covering full recipe disclosure without a mandated compute form, protocol pinning for foundation-model and zero-shot evaluations, seed and variance honesty at vision training scale, and writing results that stay checkable across the two-year gap to the next ICCV.

brycew6m878★ · +32/wk · 1 repos on radarProfile →
claude-codeMIT
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill iccv-reproducibility --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

Facts
Files in the skill folder: 1
SKILL.md size: 6 KB
Bundled scripts: none
Path: ICCV-Skills/skills/iccv-reproducibility/SKILL.md
Open the folder on GitHub →
Where it comes from
Stars: 909 · +31 this week
Language: Stata
Read our review of the source →

Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.

From the SKILL.md

# ICCV Reproducibility ICCV 2025 imposed no compute-reporting form and no reproducibility checklist that could be verified at check time (2026-07-08) — which means the venue's reproducibility bar is enforced socially: by reviewers who re-implement things for a living, and by a two-year horizon in which your paper is the standing reference until the next ICCV. Absent a form, the paper itself must carry the full disclosure. This skill is the audit. ## The two-year checkability test A CVPR paper gets superseded in twelve months; an ICCV paper's numbers get re-quoted, re-run, and re-attacked for at least twenty-four. Write every result so that a stranger in the *next* odd year can adjudicate a discrepancy: - Dataset **version and split files** named (not "standard split" — the standard moves), with checksums where licenses allow. - Metric **implementation** cited by repo and version; identically named metrics differ across codebases by more than typical paper deltas. - Pretraining **corpus and checkpoint** identified for every initialization; a gain that rides an undisclosed web-scale pretrain is a different claim than the paper makes. - Evaluation resolution, crop policy, and test-tim

What's inside
Steps it walks through
  1. The two-year checkability test
  2. Foundation-model era: pin the protocol, not just the seed
  3. Recipe as a build artifact
  4. Variance honesty at vision budgets
  5. Compute disclosure without a form
  6. Withheld test sets and server etiquette
  7. Determinism paragraph, written once
  8. Reverify each cycle
  9. Output format
More from Awesome-Journal-Skills
All skills →
About this skill
What does the iccv-reproducibility skill do?

Use when hardening the reproducibility story of an ICCV paper, covering full recipe disclosure without a mandated compute form, protocol pinning for foundation-model and zero-shot evaluations, seed and variance honesty at vision training scale, and writing results that stay checkable across the two-year gap to the next ICCV.

How do I install it?

Run `npx skills add brycewang-stanford/Awesome-Journal-Skills --skill iccv-reproducibility --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.

Where does this skill come from?

From brycewang-stanford/Awesome-Journal-Skills, a repository with 909 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.

Is a popular skill a good skill?

Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.

Keep going