Human Atlas
The human body in 3D, part by part.
In plain words
- What it is
- A 3D atlas of the human body that opens right in the browser, with nothing to install.
- Who it’s for
- Medical students, school students, teachers and anyone curious about how the body works.
- What it gives
- Turn the body, take it apart layer by layer and look at any of thousands of parts — free, in Uzbek, Russian or English.
Human Atlas is an interactive 3D anatomy atlas that runs in the browser. It shows 3,357 modelled structures from the open BodyParts3D and Z-Anatomy datasets, grouped into 17 body systems — from the skeleton and muscles to arteries, veins and the nervous system — on male and female models.
Business context
Anatomy atlases are either paid desktop software or flat pictures. Open 3D datasets exist, but they are far too heavy to send to a browser as they are.
The problem
Good 3D anatomy tools are usually paid desktop apps in English. Students in Uzbekistan rarely have an accessible way to explore the body interactively in their own language.
The solution
A browser app built with React and three.js that streams compressed model chunks progressively, so thousands of parts can be explored on an ordinary device — with search, system toggles and an exploded view, in Uzbek, Russian and English.
What I built
Built alone — architecture, code and production. No other developers worked on it.
Architecture
- A static web app: all geometry is prepared ahead of time and streamed in chunks
Frontend & apps
- React + three.js viewer: body systems and layers, structure search, exploded view, male and female models
- Three parallel loaders with progress; geometry merged per body system; shaders handle visibility and selection
- Interface in Uzbek, Russian and English
Database
- Male model: BodyParts3D 4.0 — 3,357 parts, 17 systems, 29 chunks, 3.68 M triangles (simplified from 6.68 M with meshoptimizer)
- Female model: Human Reference Atlas 1.7 — 2,914 parts, 18 systems, 5.24 M triangles
- Own binary format: Float32 positions, Int16 normals, Uint32 indices, gzip-compressed and unpacked with the browser’s DecompressionStream
Technical decisions & trade-offs
- Simplify before shipping — Meshes were reduced from 6.68 to 3.68 million triangles with quadric simplification before they reach the browser. (Trade-off: Fine detail is lost on the smallest structures.)
- A minimal binary format instead of glTF — Positions, quantized normals and indices in gzip chunks, decompressed natively by the browser — no decoder library to download. (Trade-off: A custom format to maintain; no standard tooling can open it.)
- Merge per system — Thousands of parts are merged per body system and shown or hidden in shaders, keeping draw calls low. (Trade-off: Selecting one part needs extra per-vertex data.)
Core modules
- 3,357 modelled anatomical structures
- 17 body systems with per-system toggles
- Male and female models
- Search for any structure
- Exploded view slider
- Front, side and back views
- Progressive loading of compressed model chunks
- Uzbek, Russian and English interface
- Source and attribution for open datasets
Results
- 3,357 structures in 17 systems (male) and 2,914 in 18 systems (female)
- Licences and attribution shown in the app
human.my-it.uz
Boburbek