mayalok

Turn ordinary smartphone video of an indoor space into a photorealistic, navigable 3D digital twin.

mayalok is the reconstruction pipeline behind Mayalok: it takes a phone walkthrough (video and/or photos) and produces a web-ready scene bundle — a compressed 3D Gaussian Splat (.spz), a walkable occupancy grid for obstacle-aware pathfinding, a top-down minimap, POIs and a QC report — that the Mayalok browser viewer loads for first-person walkthroughs, place search and turn-by-turn indoor navigation.

• Live viewer: https://mayalok.vercel.app
• Docs: https://the-harsh-vardhan.github.io/mayalok/
• Source: https://github.com/The-Harsh-Vardhan/mayalok

Install

pip install "mayalok[sfm]"     # includes pycolmap for Structure-from-Motion
mayalok --help

The .spz export uses @playcanvas/splat-transform (Node.js 18+; fetched automatically with npx if not installed). Training needs a 3DGS trainer such as gsplat or Brush.

Usage

# Full pipeline from a walkthrough video
mayalok build --input walkthrough.mp4 --work ./work --out ./scenes/my-venue --name "My Venue"

# Package an already-trained splat for the web viewer
mayalok process-splat --splat point_cloud.ply --cameras cameras.json --out ./scenes/my-venue --name "My Venue"

# Procedural test apartment (no capture needed)
mayalok synthetic --out ./scenes/demo-apartment

How it works

Scene bundle

Data model

Credits

Built by Team QWERTY (Ajay Yadav, Hitesh Bhatiwal, Harsh Vardhan) for the Techfest 2026-27 (IIT Bombay) Namma Space challenge. Builds on 3D Gaussian Splatting (Kerbl et al., 2023), COLMAP, Three.js and Spark.js. MIT licence.