Chronotope
Living Heritage
in Four Dimensions

Core Insight

Multi-Scale Temporal Depth

Architecture

Three Layers of Memory

Chronotope operates across three nested spatial-temporal scales, each powered by different technology, unified through a shared coordinate system.

• 🛰️

  Orbital Memory

  Nimbo Time Machine
  Monthly satellite basemaps (2019–present) provide the macro-temporal canvas. Cloud-free Sentinel imagery reveals land-use change, encroachment patterns, and environmental drift at 2.5m resolution.

  Spatial Scale

  Landscape / District

  Temporal Resolution

  Monthly cadence, 6+ years deep

• 📱

  Street Memory

  D4RT Reconstruction
  Community-captured phone video is processed through D4RT's feedforward encoder-decoder, producing full 4D point clouds with temporal correspondence in seconds. Dynamic elements—market crowds, seasonal rituals, water flow—are preserved as living geometry.

  Spatial Scale

  Building / Block / Plaza

  Temporal Resolution

  Per-capture, sub-second dynamics

• ✦

  Material Memory

  NeRF + 3D Gaussian Splatting
  NeRF provides continuous volumetric interpolation between captured states—filling temporal gaps with plausible geometry. 3DGS renders the accumulated heritage model in real-time on consumer devices, making the full 4D archive navigable by any stakeholder.

  Spatial Scale

  Surface / Detail / Ornament

  Temporal Resolution

  Continuous interpolation across all captures

Workflow

Four Phases of Heritage Practice

1. 01
   Capture
   Community as Sensor Network
   Residents walk their neighborhoods filming on smartphones. No expertise required—D4RT's feedforward architecture handles arbitrary camera motion, changing lighting, and dynamic scenes. Each walk deposits a spatiotemporal core sample into the collective archive.
   "Sovereignty begins with the right to define what is worth remembering."

   Phone video → D4RT encoder → Global Scene Representation in <2 seconds

   Automatic geo-registration into Nimbo's coordinate system via GPS metadata

   Dynamic elements (people, vehicles, ceremonies) tracked as 4D correspondences

   No cloud processing required—runs on-device or at edge
2. 02
   Accumulate
   Living Palimpsest
   Each capture is stitched into a growing 4D model of the heritage site. NeRF interpolates between temporal states while 3DGS maintains a real-time-renderable representation. The result is a navigable temporal depth—not a snapshot, but a geological record of place.
   "Heritage is not a frozen artifact. It is the accumulated pattern of living."

   D4RT point clouds registered into shared world coordinates via Umeyama alignment

   3DGS accumulates Gaussian primitives across captures for real-time rendering

   NeRF provides continuous view synthesis between temporal samples

   Nimbo satellite time-series anchors the macro context (land use, vegetation, hydrology)
3. 03
   Narrate
   Temporal Navigation as Argument
   Community stakeholders navigate the 4D archive to construct spatial narratives for planning hearings, grant applications, and cultural documentation. The plugin generates temporal transects—fly-throughs that reveal how a place has lived, breathed, and changed.
   "The most powerful argument against displacement is the visible depth of belonging."

   Temporal scrubber moves between satellite-scale years and street-scale moments

   Split-view compares any two temporal states at any spatial scale

   Community annotations pin oral histories, photographs, and stories to 4D coordinates

   Export as interactive web experience, video, or QGIS-compatible layers
4. 04
   Advocate
   Evidence of Living Heritage
   The accumulated 4D model becomes an evidentiary instrument. Change detection algorithms automatically flag encroachment, deterioration, or unauthorized modification. Community-defined heritage boundaries gain empirical backing from multi-scale temporal data.
   "When communities own the record, they own the argument."

   Automatic change detection across orbital (Nimbo NDVI/radar) and street (D4RT) scales

   Heritage impact assessment overlays proposed development onto living 4D model

   Exportable evidence packages for UNESCO, municipal planning, legal proceedings

   Open data standard ensures community retains sovereignty over collected data

Technology Integration

Why These Four, Together

Each technology solves a problem the others cannot. Their combination produces capabilities none possess independently.

• D4RT

  4D Reconstruction Engine

  What It Does

  Feedforward transformer encodes video into Global Scene Representation. Lightweight cross-attention decoder independently queries 3D position of any point in space and time.

  Why It's Essential

  200+ FPS inference. Handles dynamic scenes. Single unified decoder for depth, point clouds, tracking, and camera parameters. No iterative optimization needed.

  Zhang et al., Google DeepMind, 2025

• 3D Gaussian Splatting

  Real-Time Rendering Layer

  What It Does

  Explicit point-based scene representation using anisotropic Gaussians with learned opacity, color, and covariance. Rasterization-based rendering at 100+ FPS.

  Why It's Essential

  Web-deployable via WebGL/WebGPU. Consumer hardware compatible. Editable—communities can annotate and modify. Compact storage for archival.

  Kerbl et al., 2023; heritage applications at CIPA 2025

• Neural Radiance Fields

  Temporal Interpolation Fabric

  What It Does

  Continuous volumetric representation that maps 5D coordinates (x,y,z,θ,φ) to color and density. Extended here to 6D with temporal dimension.

  Why It's Essential

  Fills gaps between discrete captures with plausible geometry. Enables smooth temporal navigation. Provides the continuous spatial representation that regularizes 3DGS artifacts.

  Mildenhall et al., 2020; Fang et al., ICCV 2025

• Nimbo Earth Online

  Orbital-Scale Temporal Canvas

  What It Does

  Cloud-free monthly satellite basemaps from Sentinel 1/2 via Copernicus. 2.5m resolution with super-resolution. NDVI, infrared, radar layers. API via TMS/WMTS.

  Why It's Essential

  Only platform providing monthly global updates. Already has QGIS plugin architecture. European sovereignty (GAFA-free). Built for change detection.

  Kermap, Rennes, France

The D4RT Advantage

Why Feedforward 4D Changes Everything

• No Optimization Loop

  Previous approaches (COLMAP, test-time optimization) require minutes to hours per scene. D4RT's feedforward encoder processes video in a single pass. A community member's 30-second walking video becomes a 4D point cloud before they put their phone away.

• Dynamic Scene Understanding

  Heritage isn't static. Markets pulse, water flows, ceremonies unfold. Unlike MegaSaM or π³ which fail on dynamic content, D4RT's unified decoder tracks all pixels through time—the only method that captures the living dynamics of place.

• Unified Query Interface

  One decoder for depth, point clouds, camera parameters, and temporal correspondence. No ensemble of task-specific models. The query (u,v,t_src,t_tgt,t_cam) can probe any point in space and time independently—enabling the flexible multi-scale architecture Chronotope requires.

• Trivial Parallelism

  Each D4RT query is decoded independently—no inter-query dependencies. This means the occupancy-grid algorithm can efficiently track all pixels in a video with 5–15× adaptive speedup, making dense 4D reconstruction feasible on edge devices.

Design Philosophy

Technology in Service of Sovereignty