Abu Align & Train candidate v0.1.6.6

Windows x64: one online installer supplies the desktop plus pinned CUDA panorama alignment and native panorama Gaussian training components. Internet is required during setup; a compatible NVIDIA GPU/driver is required for GPU processing. Linux x64 and macOS arm64 remain CPU desktop candidates; their CUDA/Metal training parity is not claimed. Includes the interpreter, native dependencies, Manual Control Points, M54 and native spherical COLMAP. Projects stay local. Signing, notarization and independent clean-machine qualification remain stable-release gates. See docs/DISTRIBUTION_FINAL_REVIEW.md and THIRD_PARTY_NOTICES.md. The suite has mixed licenses; restricted research integrations are not automatically bundled or authorized.

Abu Align & Train v0.1.5.8

🚀 Abu Align & Train v0.1.5.8 — One-Click Single-File Installers

Download a single installation file for your operating system below:

• 🪟 Windows (10 / 11 x64): Abu_Align_and_Train_Setup_Windows_x64.exe
• 🍎 macOS (Apple Silicon & Intel): Abu_Align_and_Train_macOS_Universal.dmg
• 🐧 Linux (x86_64 AppImage): Abu_Align_and_Train_Linux_x86_64.AppImage

Full Changelog: v0.1.5.7...v0.1.5.8

Abu Align & Train v0.1.5.7

🚀 Abu Align & Train v0.1.5.7 — One-Click Single-File Installers

Download a single installation file for your operating system below:

• 🪟 Windows (10 / 11 x64): Abu_Align_and_Train_Setup_Windows_x64.exe
• 🍎 macOS (Apple Silicon & Intel): Abu_Align_and_Train_macOS_Universal.dmg
• 🐧 Linux (x86_64 AppImage): Abu_Align_and_Train_Linux_x86_64.AppImage

Full Changelog: v0.1.5.6...v0.1.5.7

Qualified Windows NVIDIA runtime candidate — 18 October 2026

Prebuilt native panorama Gaussian trainer and matching corresponding source. Built from clean LichtFeld revision 00a96b63c0ce45691d3dd9b4dcfd02069269c40c; GPL-3.0-or-later plus included NVIDIA exception and dependency terms. Windows x64, NVIDIA SM75+ and CUDA 13 compatible driver (R580+) required. Qualified on RTX 4090: native equirectangular training, finite Gaussian exports and checkpoint resume. No cubemaps. This runtime is consumed automatically by the Abu installer; these archives are not the desktop installer. Client images, scenes, checkpoints and private logs are excluded. Cold CUDA JIT may be lengthy. Linux/macOS GPU parity, surveyed accuracy and signed stable-release status are not claimed. See the manifest for scope; upstream CUDA context-release warnings occur at shutdown after successful exports.