Menu closed
Sphinx
Advanced 3D Model Comparison Platform

3DCompare

A modern web platform for analyzing, comparing, and visualizing high-fidelity 3D assets with precision and performance.

Research & Strategy

Built from operational friction.

The platform emerged from from the realization that an exact product like this does not exist yet and the ones that do are inefficient or expensive. This lead to fragmented workflows and tooling as well as inconsisten interfaces which slowed down critical engineering processes.

Challenges

The Problem

01

Fragmented Systems

Currently photogrammetry workflows and subsequent comparisons relied on disconnected tools with inconsistent workflows and duplicated operational overhead.

02

Poor Data Visibility

Scan and Comparison metrics lacked centralized visualization and actionable context.

Solutions

The Response

01

Unified Interface

A cohesive application that consolidates infrastructure tooling into one streamlined experience.

02

Real-Time Monitoring

Integrated statistical analysis provide immediate visibility into system performance and anomalies.

Try it yourself!

Interactive Demo

3D reconstruction

See how we can transform a simple smartphone video into a detailed 3D model.

Loading Model

Format

GLB

Points

0

Render

WebGL

Interaction

Orbit Controls

01

Capture a Video

Record a smooth 30–60 second video around the object, ensuring that consecutive views overlap and all visible surfaces are captured.

  • Equipment: Any smartphone camera
  • Settings: 1080p, 30fps
  • Tips: Consistent lighting works best
  • Avoid motion blur
Capture a Video
02

Frame Selection & Preprocessing

Frames are extracted from the video and filtered to keep only images that contribute new viewpoints while reducing redundancy and processing time.

  • Automatic frame sampling
  • Background removal
  • Image quality filtering
  • Reduced computational load
Frame Selection & Preprocessing
03

Feature Detection & Matching

Distinctive visual features are detected in each image and matched across multiple views, allowing the system to estimate camera positions and scene geometry.

  • Identify distinctive points on object
  • Match these features across frames
  • Estimate camera positions for each shot
  • Recover sparse scene structure
04

Sparse Reconstruction

Matched features are triangulated into a sparse 3D point cloud while camera positions are refined.

  • Structure-from-Motion
  • Sparse point cloud generation
  • Camera pose optimization
  • Initial scene geometry
Sparse Reconstruction
05

Dense Reconstruction & Cleaning

Additional depth information is calculated across all images to generate a detailed dense point cloud, which is then cleaned to remove noise and outliers.

  • Multi-view stereo reconstruction
  • Millions of reconstructed points
  • Noise filtering
  • Outlier removal
  • Explore sparse, dense, and cleaned point cloud in the viewer above
Dense Reconstruction & Cleaning
06

Surface Reconstruction

The point cloud is converted into a continuous mesh, filling the gaps between points so the result is a solid, exportable surface

  • Poisson surface reconstruction algorithm
  • Smooth, watertight mesh
  • Ready for comparison or export
Surface Reconstruction

Now let's compare two objects...

Interactive Demo

3D comparison

See how we compare the created 3D model against a reference scan to measure geometric accuracy.

Loading Model

Median Error

2.5 mm

95th Percentile Error

16.0 mm

RMSE

36.0 mm

Overlap

97.3 %

Points

0

01

Alignment

Before two scans can be compared, they need to share a coordinate system — this step rotates, scales, and positions them to match.

  • PCA-based initial alignment
  • ICP refinement for precision
  • Scaling and normalization across scans
Alignment
02

Distance Analysis

Once aligned, the nearest geometric differences between the models are calculated to measure how far each region deviates from the reference.

  • Nearest-neighbor search
  • Point-to-surface distance computation
  • Deviation measurements
  • Error statistics generation
03

Visualization & Metrics

Measured deviations are visualized as a color-coded heatmap and summarized using quantitative accuracy metrics.

  • Point-to-point distance calculation
  • Creation of color-coded deviation heatmap
  • Gathering metrics like mean deviation or RMSE
  • Switch between the reference scan, the 3D model, and the comparison in the viewer above
Visualization & Metrics
Currently in development

Target Launch Q4 2026

This project is currently evolving through active iteration, system refinement, and interface testing. Additional features, technical insights, and launch details will be shared progressively.

Thank you for reading

CONTACT

Feel free to contact me with questions, offers or ideas.