Aholo Viewer – Manycore Tech's Open 3D Gaussian Splatting Browser

1. What Is Aholo Viewer

Aholo Viewer is Manycore Tech's (群核科技) open high-performance Web renderer for 3D Gaussian Splatting (3DGS). Its chunk-level LOD streaming loads city-scale scenes with up to 1 billion Gaussians for second-level startup and smooth roaming—even on phones. It supports ply, spz, sog, splat, and related formats, with built-in LOD generation, format conversion, and collision mesh tooling—key infrastructure for Manycore's "3D internet" vision, pushing 3D from "viewable" to "usable."

Technical positioning and domain: Real-time neural rendering on the Web— not just a viewer but a developer platform for large real-world 3DGS reconstruction and interaction, filling the gap for billion-point Web performance.

Development background: From Manycore (Kujiale/酷家乐 ecosystem)—motivated by making high-fidelity large 3D accessible on consumer hardware without native apps.

Core value: Solves 3DGS scale and device limits via streaming LOD—turning "download an app" into "open a link."

Technical characteristics: Chunk-level LOD tree (vs splat-only LOD) for cache hit rate and scale; multi-precision data, frustum-priority streaming, OpenUSD hybrid cloud rendering.

2. Key Features

• Billion-Gaussian rendering: Chunk LOD + streaming for city scenes; sub-10s first paint in tests—rare on Web 3DGS today.

• Install-free multi-device: Phones, desktops, VR browsers—share like a video link.

• Format compatibility: ply, spz, sog, splat, lcc, ksplat native—minimal conversion friction.

• End-to-end toolchain: LOD generator, converters, collision generator, pick/edit helpers—no patchwork of libs.

• Voxel collision from 3DGS: Ray/capsule/ground/wall queries enable walk mode, camera collision, basic physics—beyond passive viewing.

• OpenUSD hybrid cloud rendering: Mix 3DGS + high-fidelity meshes in one frame; stream to weak clients.

• Render presets: Effect-first, performance-first, ultra-performance trade quality for FPS on low-end hardware.

3. How to Use

1. Requirements: Modern dev machine—8GB+ RAM, GPU ~GTX 1060 class; Node.js 18+ and pnpm; Windows/macOS/Linux.

2. Clone and install:

   git clone https://github.com/manycoretech/aholo-viewer
   cd aholo-viewer && pnpm install
   

3. Dev and build:

   pnpm dev    # local preview
   pnpm build  # production static assets
   

4. Integrate via npm:

   import { createViewer, SplatLoader } from '@manycore/aholo-viewer';
   const viewer = await createViewer({ container: document.getElementById('viewer-container') });
   const scene = await SplatLoader.load('path/to/your/scene.ply');
   viewer.addScene(scene);
   

5. Config tips: viewer.setViewerConfig({ renderMode: 'performance' }); generate colliders with ColliderGenerator for walk/interaction; preprocess mega-scenes with built-in LOD tools; use ultra-performance on mobile.

4. Pros and Cons

5. Comparison with Similar Tools

Selection guidance: Web tourism/exhibitions/digital twin previews → Aholo. UE-native AAA interaction → Luma UE. Algorithm R&D → Nerfstudio; Aholo for shipping large Web scenes.

6. Editor's Take

Aholo moves 3DGS from papers to billion-point Web delivery. Chunk-level LOD is architectural—not incremental—beating splat-only trees on load/render/memory vs Spark 2.0 at cited scales.

Integrated toolchain (LOD/convert/collide) is the practical win: teams focus on product, not plumbing. Clear fit for digital culture, online education, virtual expos, spatial Web teams.

MIT license + Manycore platform could make this a Web 3DGS standard if docs/community catch up.

— −0.5 for ecosystem/docs; top-tier core performance and utility.

7. Use Cases

• Digital tourism / smart city: Drone/camera captures → browser roam on phone or desktop.
• Virtual production: Real locations as 3DGS + mesh props in hybrid cloud previz.
• Games / interactive walks: Collision-enabled exploration in reconstructed streets or heritage sites.
• Embodied AI sim: InteriorGS-class semantic indoor scenes for navigation/manipulation training.

8. FAQ

Q: Free for commercial use?
A: Yes—MIT license, no royalty.

Q: Low-end PC?
A: Use performance/ultra-performance modes; integrated graphics OK for moderate scenes; huge scenes stream progressively.

Q: Supported formats?
A: ply, spz, sog, splat, lcc, ksplat; built-in converter included.

Q: Walking and collisions?
A: Run ColliderGenerator, attach to viewer physics API—see repo examples.

Q: Huge scene slow?
A: Precompute LOD chunks before deploy; runtime loads visible chunks only.

9. Project Links

• GitHub: https://github.com/manycoretech/aholo-viewer