Para-Fashion: Automatic bespoke parametric Fashion made practical

The recent rise in demand for individualized garment and the increased capacity of manufacturing offered by modern digital fabrication devices had traced the path for a novel set of semi-automatic design tools that assist the fashion designer in the effective creation and production of dresses. Despite the advent of this digital modelling pipelines, the majority of processes in the fashion industry are still inefficient and heavily dependent on manual work. Considering the size of the fashion industry, automatizing some of the main methods, might produce a beneficial impact on the overall process.  Concurrently we assisted to a significant rise in industrial applications of 3D acquisition, processing and optimization of 3D data. Recent technologies allow producing a clean representation of a real-world object or even acquire the shape of the human body. Contemporarily the development of novel shape analysis and geometry processing techniques had created a new arsenal of technologies that are capable of anticipating the physical behaviour and then optimize the design and fabrication of real cloths. However, nobody had successfully integrated all these aspects to a fully digital optimization process to create bespoke optimized dresses.


In this project, we aim at creating a pipeline that fits digital clothes into a personalized 3D scanned model of the human body and as automatically derives the optimal sewing patterns. Specifically:

  • Define a procedure which is capable of fitting the digital model of cloth into an acquired point cloud which approximate and individual's human body.
  • Optimize the sewing pattern and assist the designer in the modification of the current patch decomposition of the cloth. This step should illustrate on the fly where the tissue experiences the major deformations/tensions/compression and suggest the artist new cuts to accommodate for such geometric setup.
  • Exploit deeply trained model to build a novel software pipeline that automatically creates sewing patterns for new digital creation.

Reverse modelling using machine learning

Modern 3D sensor technologies allow ecient digital reproductions of many real-world objects. However, the produced 3D data is usually organised as a point cloud or a triangle mesh. While this data may accurately represent the shape of an object, it is not suitable for most applications beyond simple viewing. For example, production pipelines like CG animation, games, Computer-Aided-Design or Architecture require polygonal meshes to model the structure and function of the object as well as its surface shape. These compact, highly informative, 3D assets are referred to as Base Meshes. The automatic conversion of a 3D scanned real-world object to a suitable base mesh is a challenging open problem. While the reconstructed model might be very accurate , the underlying structure is signicantly dierent from that designed by the artist. Artists enrich the 3D shape with semantic features encoded in the connectivity. Creating a base mesh that effectively captures these semantic properties, and is not cluttered with irrelevant features, is known as Reverse modelling or Re-topology. Such processes convert an input 3D data into a structure that is more readable, usable, and more abundant in semantics.


This project proposes a novel approach that combines geometry processing and machine learning to automate the process of creating a high value, to for purpose 3D base mesh from an unstructured 3D point cloud. We will achieve this goal by analysing and training a knowledge base with an extensive database of manually-modelled meshes to model the connection between shape and the underlying connectivity.

Specifically, we aim to:

  • Design a novel set of 3D shape descriptors that are capable of representing and transferring mesh connectivity.
  • Design the first system that is able to deeply understand how the mesh structure is related to object shape, identity and function at multiple scales.
  • Exploit deeply trained model to build a novel software pipeline that automatically perform Re-topology.

Architectural Geometry Optimization : Construction

Architectural geometry is an active discipline that is transforming architecture and art: it allows a designer to focus more on the aesthetics relieving him from all other practical considerations and constraints regarding physical realization, such as feasibility, stability, assembly sequence or material usage, which are instead managed by an algorithm in the background.

Architectural design takes advantage of the availability of phys- ical and mathematical foundations to simulate, anticipate and fi- nally optimize the physical behavior of structures. In recent years, a lot of efforts have been focused on automating the design process of complex architectures, such as free-form vaults, roofs, and envelopes.

While most of the attention has been focused on optimizing the geometric properties of the structures little or no attention has been dedicated to how to assemble them properly or how to derive the best assembly sequence.


  • Given a self supporting structure and a set of devices we can use to support its construction we want to derive the optimal set of devices and the better assembly sequence.

  • Additionally, the problem could be extended to other kind of structures and objects which have a non trivial optimal assembly sequence.

Architectural Geometry : Active bending structures

Bending-active structures represent one of the most exciting and cheapest solutions to fabricate doubly-curved architectural surfaces. In bending-active structures, the elements are directed to a global self-equilibrating status by the elastic deformations caused by joint bending and stretching. The curvature and the internal stresses produced in the equilibrium configuration will drive the entire design process. Given a principal constituting material, the form-defining process optimizes the individual shape of each element to match the target shapes at the equilibrium configuration. This scenario makes bending-active structures not trivial to design.

The practical applications of these structures have been limited for a long time by the lack of computing tools capable of simulating and optimizing them. However, the increase of computational power and the advent of modern fabrication technologies have recently revived everyone's interest around bending-active structures. A recent trend aims at optimizing the stiffness of bidimensional panels by variating the geometry of its mesostructure. This approach discretizes the surface into a set of sub-modules which embed some predefined patterns. Once these are actively bent, the final structure at the equilibrium state results as close as possible to the target shape.


Current approaches are still very limited, as they only consider a limited amount of predefined patterns. In this project, we would like to expand the capability of those methods by:

  • Exploring different patterns, classify their mechanical properties and define a way to automatically parametrize their shape and efficiently approximate their mechanical behaviour.
  • Employ those patterns to create a framework that optimizes the placement and the shape of the different microstructures in the vocabulary to match precise deformation behaviour.
  • Use the proposed method to create complex doubly curvature surfaces.

Scalable Topology Optimisation

Recent advancement in digital fabrication has significantly changed the strategies to deliver digital shapes to reality. Additive manufacturing techniques decouple the fabrication effort from underlying shape complexity. They are capable of fabricating arbitrary complex shapes with incredible accuracy and local control of implied materials.

Such impressive, expressive power allows designing objects with complex functionalities or beautiful structural properties that were impossible to fabricate with traditional production assets. This complexity will enable us to efficiently distribute material in specific locations to make lighter and more mechanically efficient shapes. We usually refer to this process in literature as topology optimization. In a structural design problem, given geometric and physical constraints, topology optimization algorithms aim to find a shape that minimizes the weight. Decreasing weight is a crucial requirement to save on manufacturing costs to reduce the total weight, and to make the bone replacement more comfortable.


  • Defines new topology optimization techniques to create as-light-as-possible 3D printed shapes that can efficiently resist external strength. We can realize this optimization by distributing a set of Voronoi 3D cell, which automatically optimizes their shape and distributions in the volume to support the external forces.
  • Investigate scalable algorithms to support the design of surfaces with predictable structural

properties and their integration with modern fabrication pipelines. We want to pursue this goal by using an innovative approach: integrate recent advancement on shape analysis and geometry processing with physically-based analysis.

Data-Driven Hex-Meshing

The hexahedral and hex-dominant volumetric meshing of 3D shapes is a well investigated, yet still open, research topic. They seek to generate meshes with well-shaped, or box-like, elements whose outer surface closely aligns with that of the input model.

Despite multiple attempts, quality all-hex meshing remains elusive, and industrial models are still meshed using semi-manual block decomposition, a tedious and time-consuming process. Existing automatic methods for quality all-hex meshing are applicable to only a subset of inputs; while more general purposes produce inferior quality meshes or fail to capture surface features.


  • Propose a new approach which is stable and can be applied to arbitrarily complex meshes. We will use a novel tracing procedure to decompose the domain into subdomains that can be decomposed into hexahedral elements.  
  • Integrate this new approach with a data-driven model to create volumetric meshes composed by hexahedral elements using a philosophy similar to data-driven interactive quadrangulation (see

Reverse Bioprinting of cellular spheroids: Simulation and Optimization


Bioprinting is a novel technology that is rapidly changing the state-of-the-art in personalized tissue implants. This technology is destined to explode in the next years and brin

g significative benefits to society. Similarly to classical 3D printing devices, bioprinting adds layers of cellular material iteratively to shape the desired tissue. Differently from conventional 3D printing, bioprinting organizes the deposited material in a set of spheroid stem cells embedded in a hydrogel. Once the printing process is complete, nearby spheroids merge to create a single portion of biological tissue.

Unfortunately, two main constraints limit the democratization of this technology: high cost and lack of control. The price is relative to the involved stem cells, and the time needed to complete the process. The controllability is related to the unpredictable behavior of the fusion process between adjacent spheroids that might considerably alter the final shape.

In this project, we propose a novel framework that takes advantage of recent advancements in geometry processing and physically-based simulation to predict the final shape which emerges from the fusion process. Additionally, we use our simulation machinery to optimize the placement of the spheroids to match the target shape emerging from the fusion process.

Thanks to predictability and optimization, we will improve the amount of used material and the final product, reducing the cost and speeding up the entire process.


Our research project focuses on two main


  • Simulation: In this phase, we extend the methods for physically-based simulation of viscous fluids to mimic the fusion process.
  • Optimization: We will use our simulation framework to anticipate and optimize the spheroid deposition process. A final verification protocol will asset the biological functionality and the geometric accuracy of the printed tissue.