The project brought together the expertise, ideas and skills of highly specialised teams from across UCL’s Museums, the UCL Centre for Advanced Spatial Analysis (CASA) and Kagenova.
You can see a brief overview of the project in the embedded video below.
You can explore one of the captures of the project, the UCL Flaxman gallery, in copernic worlds here.
To learn more about the project you can find further details here.
]]>In a previous post we discussed how spherical CNNs can be scaled to support high-resolution input data. However, such an approach cannot support high-resolution output data, which is necessary for dense prediction tasks, such as semantic segmentation and depth estimation.
In a series of two TDS articles we discuss how we solve the open problem of scaling geometric AI techniques for spherical data to support high-resolution input and output data.
In the first TDS article, titled Geometric Deep Learning on Groups, we review the dichotomy between continuous and discrete approaches. In the second TDS article, titled Hybrid Discrete-Continuous Geometric Deep Learning, we describe how to break this dichotomy through a hybrid discrete-continuous (DISCO) method that provides both excellent performance and high computational scalability.
Our underlying geometric AI technology has now reached a maturity where it can be applied to realise the next wave of open generative AI for geometries like the sphere and 3D more generally. You can learn more on our dedicated CopernicAI website.
]]>Over recent years we’ve progressed a cutting-edge programme of fundamental research to develop geometric AI techniques for spherical data that are highly effective, while also being computationally scalable to huge, high-resolution datasets. These developments are presented in a series of research papers published in ICLR, one of the premiere deep learning conferences (see our Research page for further details).
Our underlying geometric AI techniques have now reached a maturity where they can be applied broadly and at scale.
We have recently showcased models for classical AI problems, including Scene360 for classification of 360° scenes and Recommender360 to recommend 360° imagery based on a semantic understanding of the panoramic content.
With our recent research developments, we now have the building blocks needed to realise the next wave of open generative AI for geometries like the sphere and 3D more generally. To focus on these exciting generative AI applications, CopernicAI has become its own thing, with its own online presence.
You can read more on the dedicated CopernicAI website here.
]]>Many traditional image classification systems target images of specific objects or narrow views. For 360° images an entire panoramic view is captured and the question of interest is often to classify the entire scene of the image. Scene360 provides meaningful full panoramic scene classification for precisely this task.
Check out the examples below!
For each scene we list the most likely category, as well as few other probable but slightly less likely categories (ordered by likelihood). Additionally, the mode can immediately check whether a scene is indoors or outdoors, and to measure the presence or absence of a hundred semantic attributes, e.g. the existence of a horizon in the image, of folliage, natural light.
What is more, we can explore why the model believes an image is well characterized by a given scene by creating attention heatmaps that show the most important regions of the scene that are driving predictions.
In the following images, the model is looking for a “forest path”. The center of attention clearly identifies the path and the root systems of the trees:
You can access our Scene360 model on the AWS marketplace here.
]]>The white matter of the brain, as opposed to the more commonly known grey matter, contains the nerve fibers that connect various regions of the brain. These fibers are often called the information superhighways of the brain.
Diffusion MRI is a relatively new medical imaging modality, with a great deal of active research. Recent studies have shown that mapping the white matter of the brain, and examining changes over time, can be highly useful for studying, treating and potentially predicting neurodegenerative diseases, such as Alzheimer’s.
Kagenova’s AI technology has recently been applied to recover white matter fibre directions and tracts in the brain from the raw data acquired by a diffusion MRI scanner. This work is presented in a recent article published in Computational Diffusion MRI, which you can find here.
In this article it was shown that Kagenova’s AI techniques:
]]>“… offer distinct advantages over conventional fully-connected networks at estimating scalar parameters of tissue microstructure from diffusion MRI.”
Conventional spherical CNNs are not scalable to high resolution classification tasks. In this article we overview our research on spherical scattering networks, which can be used to scale spherical deep learning to high-resolution input data typical of many practical applications, with spherical signals of many tens of megapixels and beyond.
To learn more please see the full article on Towards Data Science.
]]>The field of geometric AI, or geometric deep learning, has emerged to extend the remarkable benefits of AI to these more complex — geometric — datasets
We’ve published yet another article in Towards Data Science providing a more accessible entry into our research, explaining how we are working towards helping to democratize geometric AI, in particular for spherical 360° data.
To learn more please see the full article on Towards Data Science.
]]>In this latest article we provide a gentle introduction to geometric deep learning, focusing on high level concepts rather than technical details.
We also discuss how spherical AI for 360° data is a particular type of geometric deep learning, falling into the group category. In fact, spherical AI is the canonical example of geometric deep learning on groups, with myriad applications.
To learn more please see the full article on Towards Data Science.
]]>Firstly, our other products also feature the copernic prefix and so we wanted to unify all of our product names. Secondly, and perhaps of more interest, our copernic nomenclature is inspired by a man called Nicolaus Copernicus.
As elaborated in a previous post, Nicolaus Copernius was a 16th century astronomer who first proposed the Heliocentric model of the solar system, with the Sun — rather than Earth — at its centre. The modern Copernican Principle takes this a step further and states that there is no special place or special observer in the Universe.
These concepts reflect the ethos of all of our products, where there is no special observer position in our immersive experiences, since we allow the user to move anywhere, and our AI platform democratises geometric AI, particularly for spherical 360° data.
Look out for more copernicAI announcements in the near future, with the release of models for spherical 360° data coming very soon!
]]>To learn more please see the full article here.
]]>