Dataviz applications in web-centric VR
Background
W3Reality ultimately anticipates the VR transition (upgrading) of all websites that exist on the World Wide Web (the Web). With this grand vision in mind, we are developing a web-centric VR platform where any web developers can build and publish websites rendered as VR.
Along with the core platform development, we think we need to establish some initial proofs that, compared to traditional 2D websites, VR-based websites are actually more expressive, intuitive, and productive for humans. To verify this, we are starting with datavisualization (dataviz) apps related to 1) geographic visualization (geoviz), and 2) Machine Learning (ML) education. Both geoviz and ML are rapidly growing topics in recent years, however, we rather chose these topics for another compelling fact: ardent researchers, developers, and early adopters are in these specialized areas. We are trying to make practical contributions and, as a consequence, bootstrap W3Reality with solid products loved by the communities, and grow together.
In the following sections, we introduce our initial attempts toward such dataviz apps in VR.
Geographic Visualization

We think dataviz is a perfect area where we can demonstrate the benefits from expressiveness and intuitiveness of VR. In particular, we are currently pursuing GIS (Geographic Information Systems) applications.
In our official Twitter, we have been posting demos of our JavaScript-based libraries for GIS. Through the conversations in the GIS community, we have realized that there is room for browser-based 3D GIS applications that potentially lead to practical VR-based GIS applications in future. At last, we have finalized and shipped our first 3D GIS library (~2K LOC in ES6 JavaScript including the lib and demo app), called three-geo that we have been developing over the last three months.
https://github.com/w3reality/three-geo
three-geo is a three.js based geographic visualization library. Using three-geo, we can build satellite-textured 3D terrain models in near real-time by simply specifying GPS coordinates anywhere on the globe. We publish three-geo under the permissive MIT license following OSS initiative of W3Reality. As an accompanying demo application using the three-geo library, we introduce geo-viewer (live | source code). This app demonstrates features such as 1) on-demand 3D terrain generation, 2) camera syncing with Leaflet, 3) terrain interaction and measurement, 4) auto camera orbiting, etc.
Machine Learning Education

We have been also developing a VR-based tool for Machine Learners. It is basically a 3D visualization of Convolution Neural Network (CNN) aiming at helping ML beginners get started and understand the clear picture of how CNN works. This is one of the AI-related educational applications that we currently develop along with the idea of expanding the Web to VR.
In this very alpha development cycle, we built a rough prototype and rendered the static content of layers and the connections between them shown in the figure. We also explored the possible compositions and code structure of this prototype software. We figured out that, for the visualization part, it is better to add an infinite plane to 3D space since it is a more natural way for human eyes. A point light source was also added for the very same reason.
One of the important challenges noticed during development is how to efficiently display the links connecting the layers of the network. Because each fully-connected layer requires M x N number of lines (where M and N are image width and height in pixels, respectively), a better idea would be showing it dynamically. Also, we have implemented two different types of links. One is for fully-connected links and the other is for convolution. Tensorflow.js has been added for a simple mock-up model. At this stage, the model is not yet linked to the VR module.
We will keep focusing more on visualization. The other part left is the interaction with layers (e.g. adjust layers’ size, add/remove layers), which is a vital feature to convince newcomers to try building ML models using this application. As for active users, at the moment, we are the only one. However, we have a plan to start recruiting early users in our academic circle in order to get feedback right after the 3D layer interaction is fully implemented.

