Institute of mathematics Simion Stoilow of the Romanian Academy

The Simion Stoilow Institute of Mathematics (IMAR) is one of the flagship research institutes of the Romanian Academy, founded in 1949 and represents one of the most significant centres of the Romanian mathematics and informatics life. It has over 120 permanent researcher positions, under ranks as CS1 (research director or professor legal equivalent), CS2 (associate professor equivalent), CS3 (lecturer equivalent), as well as researcher, assistant researcher, and contract researchers, working in major areas of mathematics and computer science. The activity at IMAR focuses on advanced scientific research in mathematics, computer science, and applications. Among IMAR’s permanent staff one can find researchers with doctorates at academic research institutions like Harvard, Pierre and Marie Curie, Oxford, MIT, INRIA, Edinburgh, Michigan Ann Arbour, SUNNY, John Hopkins or Leuven.

The Computational Learning and Visual Perception Group (CLVP) at IMAR covers an entire floor of over 300m2 of rennovated space with its own operated state of the art human motion capture, human eye tracking and cluster computing equipment. The laboratory hosts 10 researchers focusing on computer vision and machine learning. Areas of particular interest include human motion analysis, the semantic segmentation of images and large scale structured prediction models. Over the past 6 years the group has also been involved in the construction of three large-scale datasets made available online together with features, novel large-scale estimation models, processing software (visualization, evaluation): (1) 3d human pose estimation (Human3.6M contains 3.6 million poses – 2 orders of magnitude larger than existing repositories, (2) human gaze in the context of action recognition in video (650,000 fixations); (3) human gaze, action recognition in static images (over 1 million fixations from 9157 images). (2), (3) are the largest eye movement datasets ever created, novel also in the use of task controls (vs. free-viewing). The laboratory has been funded by competitive national and international agencies, where over the past 6 years has secured over 5 million euro. It has ongoing collaborations with reserarchers at the University of Bonn, University of Toronto, Lund University, University of California at Berkeley, Georgia Institute of Technology, and Google Research.

Team leader:       

Team members:

  • Vlad Olaru: 3D intelligent sensors, real-time systems and high-performance computing for computer vision programs with emphasis on human pose estimation
  • Elisabeta Marinoiui: Gesture recognition, stance detection and augmented reality
  • Alin PopaComputer vision and machine learning
  • Mihai ZanfirAction recognition and human pose estimation

Objectives and tasks:

  • DE-ENIGMA Perception
    • To focus in robust, context-sensitive and real time machine analysis of facial, bodily and vocal cues in unconstrained recording conditions that are atypical and abruptly changing as caractheristic for children with autism.
      1. Environmentally robust, personalised bodily features
  • DE-ENIGMA Database collection, annotation and release
    • To collect and release of a publicly available benchmark multilingual dataset of annotated facial, bodily, vocal and verbal recordings of interactions between children with autism and their counterpartners (therapist, robot, parent).
      1. Ethical Approval
      2. DE-ENIGMA data acquisition
      3. DE-ENIGMA data annotation
      4. DE-ENIGMA database design, release and maintenance
  • DE-ENIGMA Reasoning
    • This objective addresses the problem of automatic reasoning  about behavioural patterns of the child as observed in unconstrained recordings of facial, vocal/verbal and bodily expressions, and based on the features extracted in the objective “DE-ENIGMA Perception”.
      1. Multi-modal estimation of affect, interest and stress levels
      2. Context transfer
  • DE-ENIGMA Integration and Evaluation
    • To specify the functional requirements, interactions and constraints of DE-ENIGMA: construct the integrated DE-ENIGMA platform and assess the impact DE-ENIGMA skill training compared to traditional therapies on autistic children’s socio-emotional skills.
      1. Specifications of functional requirements, interactions and constraints
      2. Construction and integration of the DE-ENIGMA platform
      3. Final evaluation of the integrated DE-ENIGMA system
  • DE-ENIGMA Dissemination, Communication and Exploitation
    • To create public, scientific and industrial awareness of the project achievements. Dissemination and exploitation activities will be performed in strict relation with all the technical activities
      1. Dissemination strategy
      2. Organisation of challenges and benchmarking
  • DE-ENIGMA Management
    • To implement all technical, financial and administrative aspects of the project plan to ensure that it is executed in fulfillment of the contract with the European Commission.
      1. Project leadership and coordination
      2. Administrative and financial management
      3. Quality assurance and risk management
      4. Preparation for review of progress

Calea Grivitei 21
010702 BUCUREST
Romania