We present DeepSketchHair , a deep learning based tool for modeling of 3D hair from 2D sketches. Given a 3D bust model as reference, our sketching system takes as input a user-drawn sketch (consisting of hair contour and a few strokes indicating the hair growing direction within a hair region), and automatically generates a 3D hair model, matching the input sketch. The key enablers of our system are three carefully designed neural networks, namely, S2ONet , which converts an input sketch to a dense 2D hair orientation field; O2VNet , which maps the 2D orientation field to a 3D vector field; and V2VNet , which updates the 3D vector field with respect to the new sketches, enabling hair editing with additional sketches in new views. All the three networks are trained with synthetic data generated from a 3D hairstyle database. We demonstrate the effectiveness and expressiveness of our tool using a variety of hairstyles and also compare our method with prior art.

/pdf/deepsketchhair-deep-sketch-based-3d-hair-modeling-3o4vrt1ie1.pdf

DeepSketchHair: Deep Sketch-Based 3D Hair Modeling

We present sketchhair, a deep learning based tool for interactive modeling of 3D hair from 2D sketches. Given a 3D bust model as reference, our sketching system takes as input a user-drawn sketch (consisting of hair contour and a few strokes indicating the hair growing direction within a hair region), and automatically generates a 3D hair model, which matches the input sketch both globally and locally. The key enablers of our system are two carefully designed neural networks, namely, S2ONet, which converts an input sketch to a dense 2D hair orientation field; and O2VNet, which maps the 2D orientation field to a 3D vector field. Our system also supports hair editing with additional sketches in new views. This is enabled by another deep neural network, V2VNet, which updates the 3D vector field with respect to the new sketches. All the three networks are trained with synthetic data generated from a 3D hairstyle database. We demonstrate the effectiveness and expressiveness of our tool using a variety of hairstyles and also compare our method with prior art.

DeepSketchHair: Deep Sketch-based 3D Hair Modeling

Constraint-based frequent sequence mining is an important and necessary task in data mining since it shows results very close to the requirements and interests of users. Most existing algorithms for performing this task are based on a traditional approach that mines patterns directly from a sequence database (SDB). However, in fact, SDBs are often very large. The algorithms thus often exhibit poor performance because the number of generated candidates and the search space are enormous, especially for low minimum support thresholds. In addition, these algorithms must read an SDB again when a constraint is changed by the user. In the context of frequently varied constraints, repeatedly scanning SDBs consume much time. To address this issue, we propose a novel approach for generating frequent sequences with various constraints from the two sets of frequent closed sequences ($$ {{\mathcal{F}}{\mathcal{C}}{\mathcal{S}}} $$) and frequent generator sequences ($$ {{\mathcal{F}}{\mathcal{G}}{\mathcal{S}}} $$), which are the concise representations of the set $$ {{\mathcal{F}}{\mathcal{S}}} $$ of all frequent sequences. The proposed approach is based on novel theoretical results that show an explicit relationship between $$ {{\mathcal{F}}{\mathcal{S}}} $$ and these two sets and have been strictly proved. The approach is then used to develop an efficient algorithm named MFS-IC for quickly generating frequent sequences with item constraints, a task that has many real-life applications. Extensive experiments on real-life and synthetic databases show that the proposed MFS-IC algorithm outperforms state-of-the-art algorithms, which directly mine frequent sequences with constraints from an SDB, in terms of runtime, memory usage and scalability.

Fast generation of sequential patterns with item constraints from concise representations

We present a method to approximate a mannequin represented as a triangle/quadrangle mesh using spheres and capsules. We formulate this 3-D approximation problem as a 2-D one, which increases its efficiency by one or even two orders of magnitude. Our algorithm first extracts cross sections from a given mannequin, approximates each cross section with circles using a variant of iterative Lloyd clustering, and finally generates capsules based on the circle sets. Compared with previous methods, our method is much faster and is able to get a tighter result with a given number of primitives. Our method can be directly applied to cloth-body collision handling in real-time cloth animation, and the simulation results by our approach are visually plausible.

Variational Mannequin Approximation Using Spheres and Capsules

The head is one of the most sun-exposed anatomical zones and its ultraviolet (UV) exposure is often underestimated. The goal of this research is to quantify the solar erythemal UV protection provided by human hair for different orientations and Solar Zenith Angles (SZA). For this purpose, a three-dimensional (3D) numeric simulation tool called SimUVEx was used to forecast the distribution of UV exposure based on geometric information and solar irradiance. The relative UV dose reduction was expressed in terms of predictive protection factor (PPF). Results showed that the effectiveness of the protection depends strongly on the orientation of the head and on the tilt of the hairs. As the SZA increases, the PPF also increases. PPFs provided by hair with a certain degree of order (parallel alignment) were generally higher by a range of about 10%–30% than the ones provided by random arrangements. A significant exposure to the scalp through the hair, even for an average hair density, was found. It confirms that hair is an important natural sunscreen in reducing transmission of UV radiation and must be considered in the prevention messages of skin cancer on the scalp, particularly in people with alopecia.

3D Modelling for Solar Erythemal UV Protection Provided by Human Hair

With the tremendous performance increase of today’s graphics technologies, visual details of digital humans in games, online virtual worlds, and virtual reality applications are becoming significantly more demanding. Hair is a vital component of a person’s identity and can provide strong cues about age, background, and even personality. More and more researchers focus on hair modeling in the fields of computer graphics and virtual reality. Traditional methods are physics-based simulation by setting different parameters. The computation is expensive, and the constructing process is non-intuitive, difficult to control. Conversely, image-based methods have the advantages of fast modeling and high fidelity. This paper surveys the state of the art in the major topics of image-based techniques for hair modeling, including single-view hair modeling, static hair modeling from multiple images, video-based dynamic hair modeling, and the editing and reusing of hair modeling results. We first summarize the single-view approaches, which can be divided into the orientation-field and data-driven-based methods. The static methods from multiple images and dynamic methods are then reviewed in Sections III and IV . In Section V , we also review the editing and reusing of hair modeling results. The future development trends and challenges of image-based methods are proposed in the end.

A Survey of Image-Based Techniques for Hair Modeling

In the past years, significant progress has been made in image-based hair modeling, thus producing abundant 3-D hair models. However, on the one hand, the reconstructed hair models could not preserve the structural details of hairstyle. On the other hand, there exists little research on these modeling results. Currently, hair geometry is mostly represented as mass chains of 3-D points. It is difficult to simulate hair directly from this representation. In this paper, we propose a novel approach to convert hair geometry model into helices, which could be easily plugged into dynamic hair simulation. We construct a hair model from a hybrid orientation field, which is generated from four fields. We extract a representative guide hair strand model from this hair geometry. Then, we use adaptive floating tangents fitting algorithm to convert this hair geometry into a physics-based hair model. To initialize this hair model, we calculate a corresponding static equilibrium configuration under external forces, including gravity, frictional contacts, and viscous drag from ambient air. We simulate dynamic hair by the Euler–Lagrange equations. Our approach can preserve structural details of 3-D hair models, and can be applied to simulate various hair geometries.

Yongtang Bao

Papers

A Survey of Image-Based Techniques for Hair Modeling

An Image-Based Hair Modeling and Dynamic Simulation Method