Vous êtes ici : Accueil » Kiosque » Annonce


Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?


21 décembre 2020

Postdoctoral position on 3D object recognition for automatic identification of benthic fauna

Catégorie : Post-doctorant


Postdoctoral position on 3D object recognition for automatic identification of benthic fauna

Duration: 24 months.

Desired hiring date: the position is open from 1 January 2021. The call will remain open until satisfactory candidate is found.

Take-home salary: 2082€/month.

Workplace: Lab-STICC UMR CNRS, Brest National School of Engineering, Brittany, France.


Context of the postdoctoral position:

The Lab-STICC is a research unit of the French national center for scientific research (CNRS). The staffs (more than 570 people) are located over the different institutes on several geographical sites in Brittany working within one central theme: “from sensor to knowledge”.

We are seeking an outstanding postdoctoral research fellow with experience in computer vision and machine learning to work on a project investigating the design of algorithms for 3D object recognition, which is part of a collaboration with EQUINOR and IFREMER.

The postdoctoral position is funded under the research project BLUE REVOLUTION supported by the Equinor Norway. The BLUE REVOLUTION project gathers research units, companies, engineering schools and universities around the design of a powerful tool based on artificial intelligence (AI) to resolve the problem of marine benthic diversity description for environmental impact assessments and biodiversity surveys. BLUE REVOLUTION is also mainly supported by EUR ISblue (Interdisciplinary Graduate School for the Blue Planet, Brest, co-funded through by ANR grant ANR-17-EURE-0015).


Objectives and challenges:

The postdoctoral research fellow shall focus on the development of an automatic taxonomic classification tool using AI and 3D fluorescence imaging for high throughput analysis. This mission aims at developing reliable 3D object recognition methods for benthic species identification by taking advantage of the maturity of 2D deep architectures and for computing requirements. Based on recent developments in 2D Convolutional Neural Networks (CNNs), the postdoc shall propose and develop an end-to-end neural architecture that allows extracting automatically the relevant features for the classification task to be performed. The first step will consist in assimilating the methods and algorithms for 2D/3D recognition proposed in the deep learning literature. Based on these methods, the postdoc shall develop effective and rapid methods for massive 3D object recognition. To address the multiple classes issue, we will focus on the aspects of incremental learning by transfer learning that allows the network to be adapted to successive classes considered as new situations. To cope with the broad size, abundance, and complexity ranges characterizing infaunal benthic organisms, the training set will be organized into a hierarchical framework, then the recognition system performance will be evaluated at different taxonomic levels. This approach will advance understanding of deep neural networks used in 3D object recognition and will bring more clarity to the decision-making mechanisms of these systems, which is still largely misunderstood.


Sub-topics/subtasks include:

- A state of the art and an understanding of the deep learning methods proposed for 2D/3D object recognition.

- Designing a deep-learning architecture trained end-to-end in order to perform massive identification of species on the most abundant groups.

- Design and implement AI algorithms for 2D object recognition.

- Design and implement AI algorithms for 3D object recognition.

- A new composite system for 3D object recognition.

- Implementation of a validation protocol with performances evaluation (in terms of size of the training set, training time, processing time, precision and recall). The functional prototype could be shared with the other partners of the project to provide first in situ results.

- Delivery of results on benthic species identification.

- Comparison performances (drawing strengths and weaknesses) between the different proposed frameworks.

- Preliminary applications in meaningful seafloor activities.

- Publishing of the results and participation in dissemination activities.

- Redaction of technical reports and code documentation.

- Participation in project monitoring.


Required qualifications and skills:

Ph.D. degree in computer science or related field with specialization in computer vision/machine learning.

Experience within deep learning frameworks is highly recommended.

Knowledge of programming languages: C++, Python, MATLAB.

Knowledge of libraries: PyTorch, TensorFlow, OpenCV, PCL, Open3D.

Proficient in English language (written and oral).

Interpersonal skills and the ability to work in a multidisciplinary team are recommended.

Taste for research activities in biological applications.


To apply:

If your interests are compatible, please feel free to send the following information in a single PDF file to abdesslam.benzinou@enib.fr and daniela.zeppilli@ifremer.fr :

- Detailed curriculum vitae, including list of publications.

- Cover letter explaining your interest and how your experience fits with this postdoctoral position.

- One-page research statement.

- Names and contact details of at least two referees.


Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2020.