Our project Freu3d is among the 18 projects selected in the European call on eXtended Reality technologies organised by XR4ALL, an initiative funded under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement no. 825545).
Freu3d aims to be a more efficient alternative to neural networks for the transformation of 2d images into three-dimensional models: it operates using the principles of Sigmund Freud’s metapsychology and the potential of p-adic numbers mathematics.
Below is the project abstract:
Freu3d aims to face two problems simultaneously: 3d reconstruction from a small amount of 2d images and implementing a new kind of AI based on an innovative and potentially groundbreaking conceptual model.
Since 3d reconstruction has become increasingly necessary for modern technologies and users, many developers are working on reconstructing objects. The reconstruction problem can be addressed in a general way with photogrammetry techniques, but this process requires several images and is not context aware. As result, this procedure does not allow to perform the abstraction process that could be useful to improve the coherence of the represented objects. Reconstruction is also being performed by using neural networks. Adopting NN it is possible only to handle specific objects and it is difficult to fix the results if something in the training has gone wrong.
We think that, in order to improve 3d reconstruction, it is necessary to introduce new paradigms on the algorithmic side. A new approach to the problem will allow us to be less reliant on neural networks and their limits.
Freu3d shows a new artificial intelligence approach. Our proposal gathers together the dynamical model of the mind, proposed by a group of scholars, based on a powerful mathematical concept (p-adic numbers) and the complex and yet actual model of the mind described by Sigmund Freud in its Metapsychology.
These two worlds share a tree-like structure that allows us to create – in a very natural way – the emergence of abstract concepts (such as categories) starting from data. Therefore, combining them in the AI model proposed by Freu3d creates the layer of abstraction useful for the reconstruction of a structure from a few photos, adopting an economic procedure from the point of view of memory consumption.