ProofRun

> This repository has moved. Active development continues at ProofRun/LoopVera. Please use the new repository for issues, contributions, and releases.
>
> 本仓库已迁移。 后续开发请前往 ProofRun/LoopVera。问题反馈、贡献与发布均在新仓库进行。

---

English

ProofRun Vision brings commercial-grade runtime debugging to open-source vision pipelines. Add observe() to your OpenCV / NumPy / PyTorch / OpenMMLab / custom code to plug into a four-pillar closed loop — See · Tweak & rerun · Diff across runs · Sign-off — without replacing any operator. Library-agnostic, replayable, and free. When debug folders fill with debug_003_v2_final.png, the gap is tooling, not algorithms. Install via pip install proofrun-vision, run once, then open the local workbench in your browser.

The closed loop

At its core is a four-pillar closed loop: capture every intermediate (See), adjust a parameter and rerun (Tweak & rerun), compare two runs side by side (Diff across runs), and sign off when stakes are high (Sign-off) — then the loop closes and you iterate again.

---

简体中文

ProofRun Vision 把「商业视觉软件级」的运行时调试带给开源视觉栈：在 OpenCV / NumPy / PyTorch / OpenMMLab / 自研算子 代码里加入 observe()，接入一条 四支柱闭环 — 看清 · 改参重跑 · 跨次对比 · 签字担责 — 挂在你现有代码上，不替代任何算子。库无关、可回放、免费。测试集掉点、文件夹堆满 debug_003_v2_final.png 时，缺的多是开发工具而非算法。pip install proofrun-vision 安装，跑一次后在浏览器打开本地工作台即可。

闭环

核心是一条 四支柱闭环：采集中间产物（看清）→ 改参并重跑（改参重跑）→ 两版并排对比（跨次对比）→ 高风险场景人工签字（签字担责）→ 回到起点，持续迭代。

---

*Run with proof.*

Collapse