An independent, unofficial source repository for camera-feature research. It presents selected, sanitized original algorithms, UI components, and offline tests. Main development remains in a separate research project. This repository does not currently provide a complete installable camera update.

X1D II and X2D II are different models. X1D II face-priority excerpts are in X1D2/; historical candidates do not establish current camera installation or acceptance.

Pixel Overclock uses pixel-shift area-array scanning and synthesis: samples from different pixel-shift positions are mapped to a denser spatial grid. The six-sample research path reconstructs a 23326 × 17498 image: 408,158,348 pixels, approximately 408 MP. Six input images count the positional samples, not a 600 MP output or a change to the sensor's physical pixel count. A four-sample reference path is retained for algorithm comparisons.

The public source provides pixel-shift scan reconstruction, batched parallel reads, compute partitions, background writes, and plain-memory pipeline components. Synthetic-input comparisons verify algorithm consistency; they do not establish an equivalent gain in optical detail. Scene motion, lens resolution, sampling alignment and processing quality still affect the result. Actual camera capture, rendering, final-file saving and album integration are outside this public build. See pixel research.

This work investigates controlled fast-sweep parameters, speed limits, transitions toward fine focus, and the associated mode/UI behavior. Historical candidates are not a universal autofocus upgrade: a speed-command multiplier does not imply the same improvement in total focus time, and lens compatibility, focus accuracy and on-camera stability require separate verification. This update supplies no firmware addresses, internal bindings or installation procedure. See the AF research scope.

The public eye-preference policy covers left/right choice, choosing the eye nearest a supplied position, delayed switching when an eye is lost, returning to a preferred eye after it reappears, and falling back to a valid face target. Missing eyes do not reuse stale coordinates or display a fabricated old eye box. This is target-selection logic, not a new face detector or an accepted camera AF controller. See eye preference.

The separate grayscale tracker retains a bounded target region and rejects stale observations; it does not claim neural face detection or identity recognition. X1D II's licensed detector is a separate module and does not establish X2D integration.

This research investigates enabling ordinary flash during electronic-shutter capture under suitable exposure conditions. It is separate from high-speed synchronization (HSS) and does not establish a measured, generally safe shutter-speed threshold. Existing dynamic flash compensation must remain intact; any additional wireless timing adjustment is a separate contribution, with zero adjustment preserving the original behavior. The public scope is a sanitized research summary and an independent offline flash UI, not an installable synchronization implementation or proof of physical flash timing. See synchronization scope and offline flash UI.

• Autofocus behavior: research into sweep/fine-focus stages, bounded parameter choices and settings interaction. Historical candidates require their own dependencies and accuracy checks; they are not a universal focus-speed claim.
• Replay and image output: full-size JPEG and embedded-preview separation, loading/memory lifetime, and display-color conversion with explicit ICC inputs. A responsive preview and full-resolution inspection are distinct goals.
• Flash and exposure timing: ordinary-flash policies, group/power settings and optional half-press updates, with original compensation responsibilities preserved. Offline policy checks do not establish physical synchronization or radio transmission.
• UI and state restoration: original animation, abstract audio routing, settings persistence and failure recovery. No audio assets, manufacturer resources or complete device installer are supplied.

See first-generation X1D and the build scope for available components and remaining dependencies.

The second-generation X1D work focuses on face-priority single autofocus (AF-S): validating detections, selecting a target, exchanging bounded target state and processing previews. Conditional eye-geometry candidates are part of the source excerpts, without an accepted on-camera eye-AF claim.

The public module includes original selection/exchange/publisher excerpts and a fixed BSD-licensed detector. The detector builds independently and passed a synthetic blank-image call; the complete publisher still lacks four project headers, so full AF integration and detection-quality acceptance are not claimed. See X1D II. This is separate from X2D II, whose current public content remains a topic index.

Use Python 3.11+ and, for these tested examples, Zig 0.13.0. Specify an output directory. These entry points do not connect to a camera.

python -B scripts/CodeTests/test_build_contract.py
python -B scripts/build_core.py --compiler zig --driver zig --build-dir ./outputs/core
python -B scripts/build_pixel_research.py --compiler zig --build-dir ./outputs/pixel-research
python -B scripts/build_x1dii_detector.py --compiler zig --build-dir ./outputs/x1dii-detector

The pixel entry checks plain-memory results against a fixed file algorithm, plus ten C component checks and eleven Python tests. The X1D II entry builds only the detector and checks a synthetic blank image; it does not build the incomplete publisher. See build methods and inputs.

Browse the grouped documentation index.

• Updates and verification status
• Scope, source review, and contributing
• Layout and stage summary

Current documents use concise topic summaries, without historical firmware addresses, device records or internal work logs. Git history has not been rewritten; documentation cleanup is not a comprehensive rights review of older source.

Original work uses MIT. The detector in X1D2/face-afs/vendor/libfacedetection/ retains its original BSD-3-Clause terms, notices and publicly released model parameters. Project licensing does not grant third-party rights. This project is not affiliated with or endorsed by the manufacturer.

独立、非官方的相机功能研究源码仓库，展示经过筛选和脱敏的原创算法、界面组件及离线测试。主要开发在独立研究项目进行；本仓库目前没有完整可安装的相机更新包。

X1D II 与 X2D II 是不同机型；二代人脸优先源码位于 X1D2/。历史候选不代表当前公开源码已经完成相机安装或功能验收。

像素超频采用像素位移面阵扫描合成：把不同位移位置取得的采样映射到更密的空间像素阵列。六合一研究路径的目标图像为 23326 × 17498，共 408,158,348 像素，约 4.08 亿像素。六张指位移采样数量，不是六亿像素输出，也不改变传感器的物理像素数量；源码还保留四个位置采样的参考路径供算法对照。

公开部分提供位移扫描重建、分批并行读取、计算分区、后台写入和通用内存流水线。自造输入对照证明算法结果一致，不等于光学细节也提高同样倍数；场景运动、镜头分辨力、采样对齐和处理质量仍会影响结果。实际相机拍摄、显影、成片保存和相册接入不在公开构建范围，见像素研究。

研究受控快扫参数、速度限制、转入精扫的阶段衔接，以及相应的模式与界面行为。历史候选不能视为通用对焦升级：速度命令倍数不等于总对焦耗时提升倍数，镜头兼容性、合焦精度和机内稳定性仍需分别验证。本次不提供固件地址、内部绑定或安装流程，见对焦研究范围。

公开的人眼偏好策略包括左右眼选择、按给定位置选择最近眼睛、丢失后的延迟切换、偏好眼重新出现后的恢复，以及回退到有效的人脸目标。眼睛丢失时不继续使用旧坐标，也不伪画上一帧眼框。这是目标选择逻辑，不是新增的人脸检测器或已验收的机内对焦控制器，见人眼偏好。

独立的灰度模板跟踪用于有限时长的目标区域保持和过期观测拒绝，不宣称神经人脸检测或身份识别。X1D II 的开源检测库属于另一个模块，不能据此认定 X2D 已完成接入。

研究目标是在合适曝光条件下，让电子快门拍摄使用普通闪光同步。它与高速同步 HSS 分开，也没有给出经测量确认、普遍可靠的快门速度阈值。原有动态闪光补偿必须保留；额外无线时序调整应独立叠加，零值保持原有行为。公开内容为脱敏研究摘要与独立离线引闪界面，不是可安装的同步实现，也不证明物理闪光时序已验收，见同步范围和离线引闪界面。

• 对焦行为：快扫／精扫阶段、受控参数与设置交互研究。历史候选仍需对应依赖与精度验证，不宣称普遍的对焦提速。
• 回放与图像输出：全尺寸 JPEG 和内嵌预览的区分、加载与内存生命周期，以及显式 ICC 输入的显示颜色转换。快速预览与全分辨率细节检查是两个目标。
• 引闪与曝光时序：普通闪光策略、分组／功率设置与可选半按更新，保留原有补偿责任。离线策略检查不证明物理同步或无线发射已完成。
• 界面与状态恢复：原创动画、抽象声音路由、设置持久化和失败恢复，不附带音频、厂商资源或完整设备安装包。

可公开组件和依赖缺口见第一代 X1D与构建范围。

重点为人脸优先单次对焦 AF-S：检测结果检查、目标选择、有界目标状态交换和预览处理。源码片段包含受条件约束的眼部几何候选，不宣称机内人眼对焦已验收。

公开模块包含项目自写的选择／交换／发布器片段，以及固定版本的 BSD 开源检测库。检测库已独立编译并通过合成空白图调用；完整发布器仍缺四个项目头文件，不宣称完整对焦接入或检测质量已验收，见 X1D II。它与 X2D 二代分开，后者当前公开内容仍为主题索引。

需要 Python 3.11+；编译示例使用已验证的 Zig 0.13.0。输出由使用者显式指定，测试不连接相机。

python -B scripts/CodeTests/test_build_contract.py
python -B scripts/build_core.py --compiler zig --driver zig --build-dir ./outputs/core
python -B scripts/build_pixel_research.py --compiler zig --build-dir ./outputs/pixel-research
python -B scripts/build_x1dii_detector.py --compiler zig --build-dir ./outputs/x1dii-detector

像素入口验证普通内存与固定文件算法的一致性，另运行十个 C 组件检查和十一项 Python 测试。X1D II 入口只编译开源检测库并测试自造空白图，不编译缺少依赖的完整发布器。详细输入见 构建方法和依赖清单。

按类别浏览文档索引。

• 本次更新与当前验证状态
• 公开范围、来源审核与贡献说明
• 目录索引与阶段成果

当前文档使用简短主题说明，移除了历史固件地址、设备记录与内部施工过程。Git 历史仍保留；既有旧源码的全面权利审核也未因此自动完成。

原创部分采用 MIT。X1D2/face-afs/vendor/libfacedetection/ 按其原始 BSD-3-Clause 许可发布，保留上游声明及公开模型参数；项目许可不覆盖第三方权利。本项目不代表厂商或获得厂商认可。