Post-Doc position 12 months in video-based gait disorders recognition with multimodal LLM

Goals:

• Optimize pose detectors (using, e.g., OpenPose, MMpose, and Theia) for the analysis of human movements in video sequences.
• Identify the main factors associated with gait disorders and the risk of falling.
• Develop specialized machine learning and computer vision models to recognize risk behaviors associated with gait and balance disorders.
• Integrate an LLM to analyze and interpret video and text data associated with gait disorders.
• Design and implement a diagnostic support platform using the technologies developed.
• Validate the tools and platform through experiments under ecological conditions.

Missions:

• Conduct collaborative research with LAMIH in France and LIVIA, LOI and IRL ILLS at ÉTS Montréal in Canada to develop innovative tools for detecting gait disorders.
• Specification of the platform, participation in its design: definition of its architecture, technological choices, etc.
• Publish research results in scientific journals and international conferences.
• Participate in writing reports and technical documents.
• Collaborate with researchers, engineers, clinicians and other partners to integrate research results into practical applications.

Applicant profil:  

• PhD in automation, computer science, biomedical engineering, biomechanics, electrical engineering or related field.
• Demonstrated experience in computer vision, machine learning and video analysis.
• Knowledge of pose detection and human motion analysis techniques.
• Familiarity with large multimodal language models (LLM) and their application.
• Programming skills (Python, C++, etc.) and use of machine learning libraries (TensorFlow, PyTorch, etc.).
• Good written and oral communication skills in English.
• Ability to work independently and as part of a team.

Salary: 2 221,30 € minimum per month (adaptable according to experience)
Application : CV, motivation letter, recommandations sent to : Mathias Blandeau (Mathias.Blandeau@uphf.fr) and Philippe Pudlo (Philippe.Pudlo@uphf.fr)