Skip to content

anything2_readme.md

Latest commit

 

History

History
82 lines (62 loc) · 3.2 KB

anything2_readme.md

File metadata and controls

82 lines (62 loc) · 3.2 KB

WebXOS Anything 2 – Three.js Point Cloud Editor

One-file zero-dependency 3D point cloud tool

Turn any JPEG image or MP4 video (first frame) into an interactive 3D point cloud using brightness-as-depth (DA3 algorithm). Edit points live, export JSON.

Features

  • Drag & drop JPEG / MP4 onto canvas
  • Instant brightness-based depth → 3D point cloud (max ~8000 points)
  • Click to select individual points
  • Change color of single point or all points
  • Delete selected point
  • Adjustable point size, depth scale, grid, auto-rotate
  • Manual camera position + zoom sliders
  • Export full point cloud as JSON
  • Real-time FPS & point counter
  • Fully offline – single HTML file

Quick Start

  1. Download or copy anything2.html
  2. Open in any modern browser (Chrome/Edge/Firefox)
  3. Drag a JPEG photo or short MP4 onto the 3D view
    → point cloud appears instantly
  4. Orbit with mouse, click points to edit

Controls

Left Panel Tabs

  • Point Cloud – upload, recompute, reset, export
  • Point Editor – select point → change color / delete
  • View Controls – point size, depth scale, grid, auto-rotate

Right Overlay (DA3 Point Cloud Controls)

  • Camera Zoom / X / Y / Z sliders for precise framing

Keyboard / Mouse

  • Left click + drag → orbit
  • Right click + drag → pan
  • Scroll → zoom
  • Click any point → selects it (info shown in Point Editor tab)

Export Format (point_cloud.json)

{
  "points": [
    { "x": 1.23, "y": -0.45, "z": 3.67, "r": 0.9, "g": 0.1, "b": 0.2 },
    ...
  ],
  "count": 5421,
  "pointSize": "0.5",
  "depthScale": "12"
}

Tips for Best Results

  • Use high-contrast photos (portraits, objects, landscapes work great
  • Bright areas = closer, dark areas = farther
  • Keep source < 4K (automatically downscaled)
  • Recompute Depth after changing Depth Scale slider

Tech

  • Three.js r128 (CDN)
  • OrbitControls
  • No backend, no build step, no npm

USE CASES:

Quantization: Reducing edge model size by using lower-precision numbers in 3D rendering (e.g., reducing models by 4-8 times with minimal accuracy loss).

Pruning: Removing unnecessary or redundant edge neural network 3D processing.

Knowledge Distillation: Training a smaller "edge" model to replicate the behavior of a larger "database" model.

Hybrid Edge-Cloud Architectures: Seamless integration where complex spacial model training happens in the powerful cloud, while real-time, low-latency inference occurs at the edge. Workloads distributed dynamically based on requirements.

Privacy and Security Focus: Processing sensitive data locally on a device inherently improves privacy and security by reducing the need for data transmission to the cloud, helping meet regulations like GDPR and HIPAA.

5G Integration: For ultra-low latency and high bandwidth 5G networks to enable new edge AI applications, particularly those requiring real-time communication, such as autonomous vehicles and remote healthcare monitoring.

Generative Computing at the Edge: Adapt Large Language Models (LLMs) and Small Language Models (SLMs) to run locally on edge devices for applications like offline translation and local voice assistants for 3D json data.

License: MIT