PhD Position F/M Hair Capture and Modeling

The Inria Grenoble research center groups together almost 600 people in 23 research teams and 7 research support departments.

Staff is present on three campuses in Grenoble, in close collaboration with other research and higher education institutions (University Grenoble Alpes, CNRS, CEA, INRAE, …), but also with key economic players in the area.

Inria Grenoble is active in the fields of high-performance computing, verification and embedded systems, modeling of the environment at multiple levels, and data science and artificial intelligence. The center is a top-level scientific institute with an extensive network of international collaborations in Europe and the rest of the world.

Topic

Capturing and modeling the dynamics of human hair is a difficult problem, still little explored by the research community. The complex structure of thin intertwined strands makes recovering hair models from images or videos a particularly challenging task.

Pioneering works consider multi-view stereo reconstructions of point clouds which are further fitted to learned examples of hair strands, as in [1]. Still building on multi-view stereo, [2] proposes to modify the approach and to reconstruct line curves instead of points. Recent works in the field investigate deep learning strategies. For instance [3] directly infers a parametric strand representation from a single view using an auto encoder architecture. In [4] a Variational Auto-Encoder (VAE) is trained to learn latent volumetric representations of hairs which combine occupancy and orientation information. The VAE is further conditioned on image input and can generate a hair representation from a single image. [5] proposes a Generative Adversarial Network (GAN) to generate 3D volumetric and orientation fields given 2D hair orientations and bust depths as extracted from a 2D image. The 3D orientation volumes can then guide hair synthesis.

All the above approaches recover only static hair representations from image information. In [6] the dynamic aspect is tackled with a strategy that builds on [5] to produce 3D orientation volumes in addition to motion encoded as 3D warping volumes.

The PhD objective is to recover dynamic hair models given image sequences. A first task is related to the representation to be considered for that purpose. Recent works in the field build on 3D orientation volumes as mentioned before. A potentially interesting direction in that respect is provided by [7] which considers dynamic garments and represent them with a coarse 2D template over which dynamic information is encoded as a texture. A similar strategy could be explored in 3D to represent hair.

The second task will consist in recovering hair models from input images and proposing a data driven approach for that purpose. A dataset of 3D hair models and corresponding image renderings will be provided. Augmenting this dataset and finding suitable representations of hair for processing will be another task of the PhD.

References:

1. Liwen Hu, Chongyang Ma, Linjie Luo, and Hao Li. Robust hair capture using simulated examples. ACM TOG 2014.
2. Strand-accurate Multi-view Hair Capture Giljoo Nam, Chenglei Wu, Min H. Kim,Yaser Sheikh, CVPR 2019.
3. Yi Zhou, Liwen Hu, Jun Xing, Weikai Chen, Han-Wei Kung, Xin Tong, and Hao Li. HairNet: Single-View Hair Reconstruction using Convolutional Neural Networks. ECCV 2018.
4. Shunsuke Saito, Liwen Hu, Chongyang Ma, Linjie Luo, and Hao Li. 2018. 3D Hair Synthesis Using Volumetric Variational Autoencoders. ACM TOG 2018.
5. Hair-GANs: Recovering 3D Hair Structure from a Single Image, Meng Zhang Youyi Zheng IEEE TVCG.
6. Dynamic Hair Modeling from Monocular Videos using Deep Neural Networks, L. Yang, Z. Shi, Y. Zheng, K. Zhou, ACM TOG 2019.
7. Dynamic Neural Garments M. Zhang, D. Ceylan, T. Wang, N. Mitra, ArXiv 2021.

See assignements.

Candidate profile

• Master in Computer Science or Applied Mathematics.
• Solid programming skills, e.g. python and/or C++.
• Solid mathematical knowledge in geometry, linear algebra and statistics.
• Experience with computer vision, deep learning and shape modeling is a plus.
• Experience with physics-based simulation is a plus.
• Good English level. French is not required.

• Subsidized meals
• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking (90 days / year) and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage under conditions

1st and 2nd year: 2 051 euros gross salary /month

3rd year: 2 158 euros gross salary / month