TiledViz: use cases & architecture

Comparing ensembles, not just individual images

Simulation produces sets of results that cannot be understood through a sequence of isolated figures alone. In climate science, astrophysics and brain imaging, analysis often requires comparing tens or hundreds of cases simultaneously.

TiledViz & VRWall retrieves results from a post-processing server and uses metadata to organise them. It supports sorting and grouping images, annotating thumbnails, selective zooming, moving rows and columns, and saving a layout for reuse.

Brain imaging: understanding cortical diversity

Patterns in cortical folds differ between individuals and emerge during development. Comparing them helps researchers study normal variability and anomalies associated with developmental conditions. NeuroSpin’s work includes building a dictionary of patterns from tens of thousands of brains.

This requires training datasets prepared and corrected by experts. Viewing many brains side by side, grouping similar conformations and synchronising their rotation supports this work and the validation of classification methods.

The BrainVISA platform, developed by NeuroSpin’s UNATI laboratory, provides cortical fold analysis tools. Its Anatomist visualisation software represents MRI data, brain meshes and sulci. Identifications can be corrected manually, while C++ and Python libraries support dedicated applications.

The BrainTwister prototype was developed with LRI at DIGITEO Moulon and Inria’s ExSitu team within DIGISCOPE. It demonstrated the value of image walls for NeuroSpin’s research and, in the longer term, the national CATI platform’s work on large cohorts.

From Wilder to Mandelbrot

BrainTwister used WildOS, a node-webkit platform developed at LRI by Michel Beaudouin-Lafon’s team. Multiple Anatomist instances displayed pairs of hemispheres on the Wilder wall, with approximately 100 million pixels across 6 metres.

Infrared interaction supported group selection and synchronised rotations, but moving elements between instances limited sorting. Freely comparing and rearranging brains is central to the scientists’ work.

At Maison de la Simulation, Mandelbrot provides 33 million pixels across 4.3 metres, with around twenty square metres of collaborative working space. The described setup uses a control workstation and tablets and differs from Wilder’s infrared and touch interaction systems.

An architecture for moving results

After WildOS and BrainTwister were installed at Saclay, TiledViz was designed as a generic application independent of WildOS, removing sorting constraints and supporting different families of simulations and experiments.

Visualisation applications run in Docker containers, exporting their output through VNC over secure tunnels. Sharing the host kernel and devices limits 3D rendering overhead. Containers also control communication ports and allow databases to be mounted read-only.

The described implementations include VNC communication on the cluster’s internal network and remote access through SSH tunnels. Early demonstrators ran up to a hundred containers on a shared-memory server, before distributing rendering across dedicated graphics nodes.

A client–server mechanism manages streams and launches remote containers as displayed elements move. Client startup is automated to maintain interaction across the full image wall.

Demonstrators in climate science and health

An initial prototype was built with LSCE and CliMAF work involving IPSL and Météo-France. It supported interactive sorting of climate output images using metadata, in a browser and during sessions in front of Mandelbrot.

Integrating Anatomist then enabled comparison of brain series from NeuroSpin databases. Both use cases share an objective: organise data visually so several specialists can examine it and discuss it together.