Title: ID-form – Face Identification Under Deformations
Funding source: FNR CORE PPP, ARTEC 3D
Principal investigator: Dr. Djamila Aouada
Researchers: Dr. Anis Kacem, Dr. Abd El Rahman Shabayek, Kseniya Cherenkova, Dr. Konstantinos Papadopoulos
Starting date/ Duration: 01/05/2018 – 36 months

Artec3D-Logo-Print-Color
logo

Automatic recognition of faces is a non-intrusive technology that has two main challenges: first, the large dynamics in the appearance of a face (pose, expression, occlusion), and second, the limitations due to the acquisition system (system noise, resolution, illumination). Both aspects make face recognition a highly non-linear problem that can quickly scale up in complexity. There are today impressive software that work well. What they lack is the dynamic aspect. Indeed, having an accurate face recognition technology can open the door to many innovative applications and revolutionize the interactions of humans with infrastructures and services. This revolution can only be possible if users are allowed to be in free motion and their faces to express their natural emotions. Being stoic and constrained to keep a straight face should no longer be a condition for a well-performing face recognition system. IDform proposes to robustly identify people from their faces in full dynamic conditions. The idea is to build on the success of today’s best performing face systems that use deep learning; however, instead of chasing the hugest datasets, the strategy is to use efficient facial models that can provide stable statistical information. The plan is to produce a robust dynamic facial recognition API using RGB-D cameras to be part of Artec3D’s products and commercialize it as a reviving tool for smarter machines. Given the new technology, the spectrum of applications is endless. It is expected to have a great socio-economic impact not only on Luxembourg but also on the international community.

Datasets

In the context of our research project, we are collecting a dynamic 3D face dataset. More information will be communicated soon.

Publications

Highway Network Block with Gates Constraints for Training Very Deep Networks
Oyedotun, OyebadeShabayek, Abd El RahmanAouada, DjamilaOttersten, Björn
in 2018 IEEE International Conference on Computer Vision and Pattern Recognition Workshop, June 18-22, 2018 (2018, June 19)

A survey on Deep Learning Advances on Different 3D Data Representations
Eman Ahmed, Alexandre Saint, Abd El Rahman Shabayek, Kseniya Cherenkova, Rig Das, Gleb Gusev, Djamila Aouada, Bjorn Ottersten
(Submitted on 4 Aug 2018 (v1), last revised 6 Apr 2019 (this version, v2)), arXiv:1808.01462v2 [cs.CV]

TEST

A dataset containing 400 real, high-resolution human scans of 200 subjects (100 males and 100 females in two poses each) with high-quality texture and their corresponding low-resolution meshes, with automatically computed ground-truth correspondences. See the following table.

TEST

A dataset containing 400 real, high-resolution human scans of 200 subjects (100 males and 100 females in two poses each) with high-quality texture and their corresponding low-resolution meshes, with automatically computed ground-truth correspondences. See the following table.

TEST

A dataset containing 400 real, high-resolution human scans of 200 subjects (100 males and 100 females in two poses each) with high-quality texture and their corresponding low-resolution meshes, with automatically computed ground-truth correspondences. See the following table.

TEST

A dataset containing 400 real, high-resolution human scans of 200 subjects (100 males and 100 females in two poses each) with high-quality texture and their corresponding low-resolution meshes, with automatically computed ground-truth correspondences. See the following table.