v0.5.0

[ooolabdev]

OOOSplat 0.5.0

中文 | English

---

中文

OOOSplat 0.5.0 重点加入自动参数优化、视频重建桥接补帧、显存感知训练和增量高清补拍，并进一步改善 Brush 显卡稳定性与 Gaussian 预览体验。

✨ 主要更新

自动优化与重建恢复

• 新增“自动优化（实验性）”，默认开启，也可在界面中关闭。
• 根据输入素材、质量档位和硬件情况自动规划抽帧、COLMAP 与 Brush 参数。
• 快速、均衡、精细档的视频初始采样目标分别为 6、8、12 FPS。
• 当视频初次重建覆盖不足时，会在预算范围内自动补充桥接画面。
• 桥接重建失败时会安全恢复到原始重建结果，不破坏已有模型。
• 图片序列继续保留全部有效图片，不应用视频桥接补帧。

显存感知的 Brush 训练

• 精细档会根据可用显存选择训练分辨率、密化配置和 Splat 数量上限。
• 明确检测到显存不足时，会自动降低一次训练配置并重试。
• 改善 Windows 单 NVIDIA 独显环境下的 Brush 选卡稳定性。
• 区分显存不足、显卡设备中断和素材读取错误，并提供更容易理解的处理建议。

增量高清补拍

已完成项目现在可以通过“高清补拍”补充更多细节：

• 支持视频和图片序列补拍。
• 复用原项目的 COLMAP 数据库、共享相机和稀疏模型。
• 只处理新增画面的特征、匹配和注册，避免完全重新计算相机空间。
• 使用全部已注册画面重新训练 Gaussian Splat。
• 支持透明 MOV 和透明 PNG 补拍素材，自动保留 Alpha 并生成 COLMAP Mask。
• 补拍会创建独立派生项目，不覆盖原项目。

补拍时请使用与原项目相同的设备、镜头、方向、分辨率和缩放倍率，并确保新旧画面具有足够重叠。

WebGPU 预览优化

• Gaussian 预览现在优先使用 WebGPU 渲染。
• 设备不支持 WebGPU或初始化失败时，会自动降级至 WebGL2。
• 调整模式、Gaussian 选择、动画预览和视频导出兼容两种渲染后端。
• 优化视角转动时的颜色与高光更新，减少明显的跳变和卡顿感。
• 改善预览地面网格和图形资源管理，提升长时间预览与重复进入预览时的稳定性。
• 扩大模型 Transform 缩放范围，更好地适配尺寸差异较大的 Gaussian 模型。

🎛️ 质量档位

档位	初始/桥接采样上限	视频工作分辨率	COLMAP	Brush
快速	6 / 9 FPS	长边最高 1600	长边 1200，最多 4096 特征	8000 次迭代，最大分辨率 1600
均衡	8 / 12 FPS	长边最高 1920	长边 1600，最多 8192 特征	15000 次迭代，最大分辨率 1920
精细	12 / 15 FPS	根据显存使用 3200、3840 或原生分辨率	长边 3200，最多 16384 特征	30000 次迭代，依据显存选择训练配置

关闭自动优化后，应用继续使用旧版固定比例抽帧和分辨率策略。

🔒 本地计算与隐私

OOOSplat 的画面准备、相机重建和 Gaussian Splat 训练均在本机完成。原始视频、图片、工程文件、模型和日志无需上传至云端重建服务。

本版本增加匿名 Planner 效果指标，用于评估自动优化策略，但不会上传视频、图片、PLY、项目路径、项目名称、日志或命令输出。匿名统计可在“设置 → 隐私”中随时关闭。

💻 平台支持

• Windows 11 x64：内置 CUDA 版 COLMAP，满足条件时自动使用 NVIDIA GPU，否则回退 CPU。
• macOS 15+ Apple Silicon Alpha：支持 M1 及更新芯片，内置 arm64 FFmpeg、COLMAP 和 Brush；当前版本未签名、未公证。
• Ubuntu 24.04 x86_64 Alpha：使用系统 FFmpeg、FFprobe 和 CPU COLMAP，安装包内置固定版本 Brush。

📦 下载

请根据系统下载对应文件：

• Windows：OOOSplat-0.5.0-x64-windows.exe
• Apple Silicon macOS：OOOSplat-0.5.0-arm64-macos.dmg
• Ubuntu 24.04：OOOSplat-0.5.0-x64-linux.deb

⚠️ 注意事项

• 自动优化目前仍属于实验性功能，可在界面中关闭。
• macOS 与 Ubuntu 版本目前为 Alpha。
• 超过 1 GB 的超大型 PLY 在部分 WebView2 环境中仍可能出现较高内存占用，我们会继续优化。
• 建议开始任务前预留足够磁盘空间，并更新显卡驱动。

---

English

OOOSplat 0.5.0 introduces automatic parameter optimization, bridge-frame recovery for video reconstruction, VRAM-aware training, and incremental high-resolution reshoots. It also improves Brush GPU stability and the Gaussian preview experience.

✨ Highlights

Auto Optimize and Reconstruction Recovery

• Added experimental Auto Optimize, enabled by default and optional.
• Automatically plans extraction, COLMAP, and Brush parameters from the source media, quality preset, and available hardware.
• Fast, Balanced, and Detailed start from target video sampling rates of 6, 8, and 12 FPS.
• When the initial video reconstruction has insufficient coverage, OOOSplat can add bridge frames within the remaining budget.
• If bridge-frame recovery fails, OOOSplat safely returns to the original reconstruction.
• Image sequences continue to keep every valid image and do not use video bridge-frame recovery.

VRAM-aware Brush Training

• Detailed quality selects its training resolution, densification profile, and Splat limit from available VRAM.
• After an explicit out-of-memory failure, OOOSplat can retry once with a safer training profile.
• Improved Brush adapter selection on Windows systems with a single NVIDIA discrete GPU.
• Better distinction between insufficient GPU memory, graphics-device interruption, and source-data read failures.
• Error dialogs now provide clearer recovery guidance.

Incremental High-resolution Reshoots

Completed projects can now receive additional high-resolution captures:

• Supports video and image-sequence reshoots.
• Reuses the original COLMAP database, shared camera, and sparse reconstruction.
• Processes only the new image features, matches, and registrations instead of rebuilding the complete camera space.
• Retrains the Gaussian Splat from all registered images.
• Supports transparent MOV and PNG reshoot media with preserved Alpha and automatically generated COLMAP masks.
• Creates a separate derived project without overwriting the source project.

Use the same device, lens, orientation, resolution, and zoom as the original project, and maintain sufficient overlap between the original and new captures.

WebGPU Preview Improvements

• Gaussian preview now prefers WebGPU rendering.
• Automatically falls back to WebGL2 when WebGPU is unavailable or initialization fails.
• Adjust mode, Gaussian selection, animation preview, and video export work with both rendering backends.
• Improved color and highlight updates during camera movement, reducing visible stepping and stuttering.
• Improved ground-grid and graphics-resource management for greater stability during long sessions and repeated preview entry.
• Expanded the model Transform scale range for unusually large or small Gaussian models.

🎛️ Quality Presets

Preset	Initial/bridge ceiling	Video working resolution	COLMAP	Brush
Fast	6 / 9 FPS	Up to 1600 px long edge	1200 px, up to 4096 features	8000 iterations, maximum resolution 1600
Balanced	8 / 12 FPS	Up to 1920 px long edge	1600 px, up to 8192 features	15000 iterations, maximum resolution 1920
Detailed	12 / 15 FPS	3200, 3840, or native resolution depending on VRAM	3200 px, up to 16384 features	30000 iterations with a VRAM-selected profile

Disabling Auto Optimize restores the legacy fixed-ratio extraction and resolution strategy.

🔒 Local Processing and Privacy

Image preparation, camera reconstruction, and Gaussian Splat training run locally. Original videos, images, project data, models, and logs do not need to be uploaded to a cloud reconstruction service.

This release adds anonymous Planner effectiveness metrics to evaluate automatic optimization. OOOSplat does not upload videos, images, PLY files, project paths, project names, logs, or command output. Anonymous statistics can be disabled at any time under Settings → Privacy.

💻 Platform Support

• Windows 11 x64: Bundles CUDA-enabled COLMAP. A compatible NVIDIA GPU is used automatically; otherwise COLMAP falls back to CPU.
• macOS 15+ Apple Silicon Alpha: Supports M1 and newer Macs with bundled arm64 FFmpeg, COLMAP, and Brush. The current build is unsigned and unnotarized.
• Ubuntu 24.04 x86_64 Alpha: Uses system FFmpeg, FFprobe, and CPU COLMAP, with a pinned Brush runtime included in the package.

📦 Downloads

Choose the package for your platform:

• Windows: OOOSplat-0.5.0-x64-windows.exe
• Apple Silicon macOS: OOOSplat-0.5.0-arm64-macos.dmg
• Ubuntu 24.04: OOOSplat-0.5.0-x64-linux.deb

⚠️ Notes

• Auto Optimize remains experimental and can be disabled.
• macOS and Ubuntu releases remain Alpha.
• PLY files larger than 1 GB may still cause high memory usage in some WebView2 environments. Further optimization is planned.
• Keep sufficient free disk space available and use an up-to-date graphics driver.

For complete documentation and usage instructions, see the project README.

---

This discussion was created from the release v0.5.0.