Wong Chih Chiang

Play through the three inspection stages a real board actually passes through — SPI (solder-paste print, before parts are placed), AOI (optical inspection of every visible joint), and AXI (X-ray inspection for joints hidden under BGA packages, invisible to any camera). In Benchmark mode, a small object-detection model — trained on the game's own self-labelled boards and compressed to run entirely client-side via ONNX Runtime Web — plays the AOI ladder on its own, live in your browser; you can tap in to catch whatever it misses and see exactly where its detection breaks down.

游戏还原了一块电路板真正会经历的三道检测工序 —— SPI(锡膏印刷检测,元件贴装前进行)、AOI(对每个可见焊点的光学检测),以及 AXI(针对隐藏在 BGA 封装下、任何摄像头都无法看到的焊点的 X 光检测)。在 Benchmark 模式中,一个基于游戏自动生成的标注数据训练、并压缩至可通过 ONNX Runtime Web 在浏览器端完整运行的小型目标检测模型会自行挑战 AOI 关卡 —— 当它漏检时,你可以介入补上,直接看清它的检测极限究竟在哪里。

• Three real inspection stationsSPI checks the solder-paste print before any part is placed; AOI scans every joint a camera can see; AXI X-rays the joints hidden under BGA packages. Same defect taxonomy a real line runs on.
• Boards that label themselvesEvery board is generated from a seed and carries pixel-accurate ground-truth boxes as a by-product, so the game doubles as an unlimited, human-free training set.
• Benchmark mode: ML live in your browserA small object-detection model trained on thousands of self-labelled boards, compressed to a few megabytes and run entirely client-side via ONNX Runtime Web. No server round trip.
• Honest about the gapDifficulty is one dial — subtlety — and it makes defects harder for the model too. When it misses, you tap in, and see exactly where machine vision breaks down and why a person still stands at a real AOI station.
• Daily board, leaderboards, field guideSame seed, same board on every device, so everyone can compare scores on the day's challenge. All-time leaderboards per mode, and an in-app field guide with a sample of every defect.
• Free, installable, offline-firstAdd it to your home screen and play without a connection. The detector runs on-device, so no board images ever leave your browser.

• 三道真实检测工序SPI 在任何元件贴装前检查锡膏印刷;AOI 扫描摄像头可见的每个焊点;AXI 用 X 光检测隐藏在 BGA 封装下的焊点。采用与真实产线相同的瑕疵分类体系。
• 自带标注的电路板每块板由种子生成,并作为副产物附带像素级精确的真值标注框,因此游戏本身就是一套无需人工、可无限扩充的训练集。
• Benchmark 模式:浏览器内实时运行的 ML一个基于数千块自标注电路板训练的小型目标检测模型,压缩至几 MB 并通过 ONNX Runtime Web 完全在客户端运行,无需服务器往返。
• 坦然面对检测盲区难度只有一个变量 —— 细微程度 —— 它同样会让模型更难检出。当模型漏检时你可以介入,亲眼看到机器视觉在哪里失效,以及为什么真实 AOI 工位仍然需要人。
• 每日挑战、排行榜与图鉴同一种子在所有设备上生成同一块板,人人都能比较当日成绩;每个模式都有全时段排行榜;内置图鉴为每种瑕疵配有样例。
• 免费、可安装、离线优先添加到主屏幕后无需网络即可游玩;检测模型在设备端运行,电路板图像不会离开你的浏览器。

Most trackers fail because logging is a chore; here the whole loop is photo → result → log, so the record actually gets kept. Everything else builds on that record: a calendar of every meal (with retroactive entry for the ones you forgot), one-tap hydration and protein logging, 7-day / 6-week / 6-month trend charts against your goal, a macro view that flags anything over your personal daily-value limits, and an Analyze button where the AI reads your own history and writes plain-language suggestions. Export any date range as a PDF report to share with a coach or doctor.

大多数饮食追踪应用之所以半途而废,是因为记录本身太繁琐;而在这里,整个流程只有"拍照 → 结果 → 记录"三步,记录才真正能坚持下来。其余功能都建立在这份记录之上:按日历回看每一餐(忘了记的可以补录到过去的日期)、一键记录饮水与蛋白质、7 天 / 6 周 / 6 个月的趋势图并对照目标线、以百分比显示各项营养素相对个人每日参考值并标出超标项,以及「分析」按钮 —— 由 AI 阅读你自己的记录,用通俗语言给出总结与建议。还可将任意日期范围导出为 PDF 营养报告,方便分享给教练或医生。

• AI food scanCamera or gallery, auto-compressed before upload. Calories, six nutrients, a per-item breakdown and a health tip. Recognises when a photo isn't food and says so instead of guessing.
• Water & protein in one tap150 / 250 / 500 / 750 ml hydration buttons and 1 / 5 / 10 / 20 g protein shortcuts, each with a progress bar toward your daily goal.
• Trends & AI insightsCalorie and water charts over 7 days, 6 weeks or 6 months with a goal line; macros as % of daily value with over-limit flags; an AI-written review of your own data.
• History & PDF reportsCalendar with dots on logged days, per-meal detail, retroactive logging, and PDF export for any date range — emailed to you or downloaded directly.
• Goals that fit youAge, weight, height, sex and activity level produce a suggested calorie goal (Mifflin-St Jeor) — suggested, never forced. Calorie, water and per-macro targets are all editable.
• Three languages, install anywhereEnglish, 简体中文 and 繁體中文 across the UI and the AI's replies. A progressive web app that installs to the home screen on iOS and Android, with an offline app shell and push reminders for meals, water and a daily summary.

• AI 食物识别拍照或从相册选取,上传前自动压缩。返回卡路里、六项营养素、逐项食材明细与健康建议;若照片并非食物,会如实说明而不会胡乱猜测。
• 一键记录饮水与蛋白质150 / 250 / 500 / 750 ml 饮水按钮与 1 / 5 / 10 / 20 g 蛋白质快捷键,各自配有朝向每日目标的进度条。
• 趋势与 AI 洞察7 天 / 6 周 / 6 个月的卡路里与饮水图表并附目标线;各营养素按每日参考值百分比显示并标出超标项;由 AI 基于你自己的数据撰写分析。
• 历史记录与 PDF 报告日历上以圆点标记已记录的日子,可查看每餐详情、补录过往日期,并将任意日期范围导出为 PDF —— 发送到邮箱或直接下载。
• 贴合个人的目标根据年龄、体重、身高、性别与活动量给出建议卡路里目标(Mifflin-St Jeor 公式)—— 仅作建议,绝不强制套用。卡路里、饮水与各营养素目标均可自由修改。
• 三种语言,随处可安装英文、简体中文、繁體中文覆盖界面与 AI 回复。作为渐进式网页应用,可安装到 iOS 与 Android 主屏幕,支持离线外壳缓存与用餐、饮水、每日总结的推送提醒。

Free accounts include 5 scans a day. Photos are analysed through Drevoiq's own server, so no AI keys ever reach your device. Nutrition values are AI-generated estimates, not medical or dietary advice.

免费账户每天可使用 5 次扫描。照片通过 Drevoiq 自有服务器进行分析,AI 密钥绝不会下发到你的设备。营养数值为 AI 生成的估算,不构成医疗或饮食建议。