Skip to content

v0.7.0

Pre-release
Pre-release

Choose a tag to compare

View all tags
@ooolabdev ooolabdev released this 19 Oct 15:45
· 1 commit to main since this release
0.7.0
e2af865

OOOSplat 0.7.0

中文 | English

中文

OOOSplat 0.7.0 加入全景视频与图片生成、外部模型导入、多格式导出、水平 WASD 漫游和自动定位视角，并改进图片匹配、生成进度与相机控制。

✨ 主要更新

全景视频与图片生成

  • 支持已拼接、单目、完整 360°×180°、2:1 ERP 全景 MP4。
  • 支持全景图片文件夹：至少两张同尺寸 JPG、JPEG 或 PNG。
  • 自动检测素材类型；无法确定时，在素材输入框内选择普通或全景类型。
  • 保留完整全景画面进行相机重建，仅将已注册画面转换为六面训练图片。
  • 全景使用独立训练配方，自动优化关闭时，快速／均衡／精细的 Splat 上限分别为100万／200万／300万；开启后根据档位和显存确定，最高允许1,000万。

Splat 上限不代表最终点数，也不保证显存足够。首版不支持双鱼眼、双目、VR180、裁切全景、厂商私有原片及全景高清补拍。

外部模型导入

  • 在“已完成”分组点击“导入”，选择模型文件或文件夹。
  • 支持标准／压缩 PLY、SOG、SPZ、SPLAT 和 KSPLAT。
  • 自动识别关联的 COLMAP 数据，满足条件时可创建高清补拍任务。
  • 导入后可使用现有预览、编辑和导出功能。
  • 外部文件保持只读，删除导入项目不会删除外部原文件。

不支持仅包含 COLMAP 的项目、普通点云 PLY、2DGS、LCC/LCC2 或 Streamed SOG。

多格式导出

  • “导出”菜单直接提供标准 PLY、压缩 PLY、SPZ v4 和单文件 SOG。
  • 每种格式附带简短说明，点击后直接启动，无需第二次选择格式。
  • 移除完整中转 PLY，按需处理删除、裁切、变换和 SH 旋转。
  • 导出弹窗显示进度、结果大小和文件位置，支持取消与失败重试。
  • 支持横竖屏视频和单文件离线 HTML。
  • 原始 final.ply 不变，失败或取消时保留已有导出。

SOG、SPZ 通常更小，实际体积取决于模型，压缩可能影响细节。压缩内存超限时请使用标准 PLY；当前 SPZ 编码器还存在约 ±2048 场景单位的坐标范围限制。

预览导航与相机修复

  • 新增水平 WASD 漫游，方向随相机水平朝向变化，不再随俯仰升降。
  • 漫游速度随缩放调整，支持 Shift 三倍加速，仅在透视视图生效。
  • 新增“自动定位视角”，返回首张有效训练图片的相机视角；全景使用最早已注册画面的 front 面。
  • 从三视图点击自动定位，也会直接切换至训练参考透视视角。
  • 点击自动定位后即可直接使用 WASD，无需再次点击画布。
  • 修复连续放大后右键平移过快、三视图旋转跳变及正交网格裁剪问题。

没有可靠训练相机的项目会禁用自动定位，不会用猜测视角代替。

图片顺序确认

  • 普通和全景图片超过100张时，生成前询问是否已按拍摄顺序排列。
  • 已排序使用顺序匹配；未排序或100张及以下使用穷举匹配。
  • 顺序选择会保存到任务中，重试和续跑不再询问。
  • 自动队列等待确认，不跳过当前任务；取消确认会关闭队列。

建议文件名顺序与拍摄顺序一致，并确保相邻图片有足够重叠。

检测、进度与兼容性

  • 素材类型检测使用轻量元数据读取，切换质量档位不再重复检测。
  • 改进多轮特征提取、匹配和重建的进度展示，区分真实计数与阶段完成率。
  • 全景六面准备与实际训练统一归入“Splat 训练”，不再将六面图片数当作训练步数。
  • 改进运行监控状态更新，补充匹配活动和长时间无输出提示。
  • 为已确认受影响的 PNG/XMP 分数焦距读取增加临时兼容补丁，不修改图片或元数据。
  • 修复 Rust 1.99 Clippy 检查导致的 CI 失败。

💻 平台支持

  • Windows 11 x64
  • macOS 15+ Apple Silicon Alpha：M1 及更新芯片，未签名、未公证。
  • Ubuntu 24.04 x86_64 Alpha

COLMAP 可按平台和设备条件使用 GPU 或 CPU；Brush 训练仍需要可用的图形处理后端。

📦 安装包

  • Windows：OOOSplat-0.7.0-x64-windows.exe
  • Apple Silicon macOS：OOOSplat-0.7.0-arm64-macos.dmg
  • Ubuntu 24.04：OOOSplat-0.7.0-x64-linux.deb

English

OOOSplat 0.7.0 adds panorama video and image generation, external model import, multi-format export, horizontal WASD navigation, and Auto locate view, with improvements to image matching, progress reporting, and camera controls.

✨ Highlights

Panorama Video and Images

  • Supports stitched, monoscopic, full 360°×180°, 2:1 ERP MP4.
  • Accepts folders containing at least two same-sized JPG, JPEG, or PNG panoramas.
  • Detects media types automatically; uncertain inputs require an ordinary/panorama choice in the media box.
  • Reconstructs full panoramas before converting registered images into six-face training data.
  • Uses dedicated training recipes. With Auto Optimize disabled, Fast / Balanced / Detailed cap splats at 1 / 2 / 3 million. With it enabled, quality and VRAM determine the cap, up to 10 million.

Caps do not guarantee achieved counts or sufficient VRAM. Dual-fisheye, stereo, VR180, cropped panoramas, private camera originals, and panorama reshoots are unsupported.

External Model Import

  • Import a file or folder from the Completed group.
  • Supports standard/compressed PLY, SOG, SPZ, SPLAT, and KSPLAT.
  • Detects associated COLMAP data and enables eligible reshoot tasks.
  • Reuses existing preview, editing, and export tools.
  • External files remain read-only and are not deleted when removing the imported project.

COLMAP-only projects, ordinary point-cloud PLY, 2DGS, LCC/LCC2, and Streamed SOG are unsupported.

Multi-format Export

  • Choose standard PLY, compressed PLY, SPZ v4, or bundled SOG directly from Export.
  • Format descriptions appear in the menu, without a second selection dialog.
  • Removes the complete intermediate PLY and processes edits on demand.
  • Shows progress, output size, and location, with cancellation and retry support.
  • Supports portrait/landscape video and standalone offline HTML.
  • Preserves the original final.ply and existing exports on failure or cancellation.

SOG and SPZ are typically smaller, but size varies and compression may affect detail. Use standard PLY when compression exceeds the memory budget. The current SPZ encoder also has a coordinate range of approximately ±2048 scene units.

Preview Navigation

  • Horizontal WASD movement follows camera heading without changing elevation with pitch.
  • Zoom-relative speed, Shift ×3 acceleration, perspective views only.
  • Auto locate view restores the first valid training camera; panorama projects use the earliest registered image's front face.
  • Locating from a fixed orthographic view returns directly to the reference perspective view.
  • WASD works immediately after locating, without another canvas click.
  • Fixes excessive post-zoom panning speed, fixed-view rotation jumps, and orthographic grid clipping.

Locating is disabled when no reliable training camera is available.

Image-order Confirmation

  • More than 100 ordinary or panorama images require capture-order confirmation before starting.
  • Ordered sets use sequential matching; unordered sets and sets of up to 100 images use exhaustive matching.
  • The strategy persists for retries and resumes.
  • Automatic execution waits for confirmation; cancelling disables the queue without skipping the task.

Keep filename order consistent with capture order and provide sufficient overlap between neighboring images.

Detection, Progress, and Compatibility

  • Lightweight metadata detection is reused when changing quality.
  • Improves multi-pass COLMAP progress and distinguishes result counts from completion percentages.
  • Keeps panorama preparation and actual training under Splat Training, without treating face counts as training iterations.
  • Improves runtime-monitor updates and matching activity/silence messages.
  • Adds a narrowly scoped temporary workaround for the confirmed PNG/XMP rational focal-length parsing issue, without changing image bytes or metadata.
  • Fixes CI failures caused by Rust 1.99 Clippy checks.

💻 Platforms

  • Windows 11 x64
  • macOS 15+ Apple Silicon Alpha: M1 or newer; unsigned and unnotarized.
  • Ubuntu 24.04 x86_64 Alpha

COLMAP uses GPU or CPU as supported. Brush training still requires an available graphics backend.

📦 Packages

  • Windows: OOOSplat-0.7.0-x64-windows.exe
  • Apple Silicon macOS: OOOSplat-0.7.0-arm64-macos.dmg
  • Ubuntu 24.04: OOOSplat-0.7.0-x64-linux.deb
Assets 5
Loading
0 Join discussion