• Full Time
  • Paris

Website Conservatoire national des arts et métiers

French public higher education and research institution

We are offering a fully funded 3-year position as a PhD student working on domain adaptation and semi-supervised learning for remote sensing imagery. We are looking for strong MSc. students or engineering graduates with a background in computer vision or machine learning, willing to help us push further the state of the art in semantic segmentation of aerial and satellite images.

This doctoral position aims to investigate semi-supervised training of deep models with a focus on two frequent usecases in Earth Observation. More precisely, we assume that the dataset can be split in two subsets:
• a large amount of unlabeled aerial or satellite observations,
• a small number of labeled images (“gold data”).

These labeled images come from either real acquisitions annotated by experts or from synthetic observations obtained by simulation. In both cases, the problem is the misalignment between the distribution of the labeled data and the target geographical area. For example, how can the model map accurately flooded towns in India if the only available labels come from urban areas in Europe? A possibility is to leverage simulated images. Yet, this introduces a significant perceptual bias since simulations do not accurately model specificities of real acquisitions (atmospheric perturbations, sensor noise).
This PhD will investigate two research topics in deep learning, with a focus on mapping applications, i.e. semantic segmentation of remote sensing images for land cover or land use mapping. Other applications such as height estimation or change detection might also be considered depending on opportunities.


The ideal applicant has the following qualifications:
• holds a master degree or an engineering degree in computer science, with a specialization in machine learning or computer vision,
• has some experience with deep learning (project, courses, publication…),
• is familiar with the Python programming language and at least one deep learning framework (PyTorch, TensorFlow, JAX, …),
• oral and written communication skills. French is not required but can help with the everyday life of the PhD candidate.
Although not required, an experience or a will to learn about remote sensing, interpretation of aerial and satellite images and geographical applications (climate change, disaster management) is a plus. Since motivation is the main factor for a successful thesis, all applications, even those that do not entirely fit the described profile, will be considered.

Where you will work

The Center for research and studies in computer science and communications (Cédric) is the computer science laboratory of the Conservatoire national des arts et métiers (Cnam), a prestigious French higher education institute. It is comprised of 80 faculty members and researchers, for a total of more than 180 people including postdoctoral fellows and PhD students. Its eight teams cover most areas in computer science, from data science to interactive media, discrete optimization, telecommunications and the Internet of Things. The new hire will join the Complex Data, Machine Learning and Representations team. Their research will be performed inside the MAGE project under its principal investigator Dr. Nicolas Audebert.

Organization: PhD thesis in France are done under a fixed-term work contract of 3 years (36 months, “CDD”). It is a full-time position of 35 hours/week. The expected salary is about ≈1800€ net/month. The starting date is expected at the earliest on the November 1st 2023 and at the latest on January 1st 2023, depending on the applicant.

Location: the laboratory is located in the heart of Paris, in the third “arrondissement”, at 2 rue Conté (subway “Arts
& Métiers”, lines 3 and 11).

Hiring: the application process is done in two steps: first a short half-hour interview by phone or videocall, then a longer technical interview of about one hour.

To apply: send a resume to nicolas.audebert@cnam.fr

• 44 days of paid holidays
• on-site subsidized restaurant
• partial remote work is possible
• employees’ association (music classes, on-site gym…)
• 44 days of paid holidays
• on-site subsidized restaurant
• partial remote work is possible
• employees’ association (music classes, on-site gym…)


To apply for this job please visit geo-mage.github.io.