Phd Position F - M Foundation Model For Plant Phenotyping H/F

INRIA

PhD Position F/M Foundation Model for Plant Phenotyping
Le descriptif de l'offre ci-dessous est en Anglais
Type de contrat : CDD

Niveau de diplôme exigé : Bac +5 ou équivalent

Fonction : Doctorant

A propos du centre ou de la direction fonctionnelle

Inria is the French National Institute for Research in Digital Science, of which the Inria Côte d'Azur University Center is a part. With strong expertise in computer science and applied mathematics, the research projects of the Inria Côte d'Azur University Center cover all aspects of digital science and technology and generate innovation. Based mainly in Sophia Antipolis, but also in Nice and Montpellier, it brings together 47 research teams and nine support services. It is active in the fields of artificial intelligence, data science, IT system security, robotics, network engineering, natural risk prevention, ecological transition, digital biology, computational neuroscience, health data, and more. The Inria Center at Université Côte d'Azur is a major player in terms of scientific excellence, thanks to the results it has achieved and its collaborations at both European and international level.

Contexte et atouts du poste

The successful candidate will join AgriScienceFM, an EU-funded project led by Wageningen University (the Netherlands) and dedicated to transforming agricultural research through the development of Foundation Models (FMs). While AI has made strides in plant sensing and satellite monitoring, its full potential in agriculture is often hindered by fragmented data and the sheer complexity of biological systems.

AgriScienceFM addresses these challenges by framing agriculture through the GxExM continuum: the interaction between Genetics (G), Environment (E), and human Management (M).

Our Ambition

The project's mission is to move beyond small-scale, "data-hungry" AI models that struggle to generalize. Instead, we are building three domain-specific Foundation Models trained on massive, multi-modal datasets using self-supervised learning:

- AgriFM-E (Environment): Focused on climate, soil, and satellite data to understand environmental drivers.
- AgriFM-M (Management): A document-understanding model capturing agronomic knowledge and decision-making.
- AgriFM-G (Genetics/Growth): A visual model trained on diverse plant imaging to master plant structure and phenotyping.

By creating these "jumpboards," the project aims to lower the data barriers for researchers, allowing for more accurate and adaptable AI tools in areas like crop breeding for climate resilience, soil health advisory, and precision pest management.

The PhD candidate will be hosted at Inria Montpellier, within a vibrant research environment at the intersection of computer science, geospatial data analysis, and plant biology. Montpellier is one of France's leading scientific hubs for agricultural and environmental research, home to major institutions including CIRAD, INRAE, and the University of Montpellier. The successful candidate will benefit from strong collaborative networks within the AgriScienceFM project consortium, access to large-scale agricultural datasets, and access to Inria's high-performance computing resources. Regular exchanges with European partners involved in the project will provide broad exposure to both fundamental and applied research challenges.

Mission confiée

Project objectives

This PhD project will focus on developing AgriFM-G, a Foundation Model for plant phenotyping. AgriFM-G captures plant structure and phenotypic variation through visual and multi-modal data, including:

- Field-level RGB imagery at the plant or plot scale.
- UAV-acquired RGB imagery.
- Field sensor readings and time-series trait datasets.

The model will be trained to encode fine-grained plant traits and phenotypic variability across diverse crop growth stages. AgriFM-G will support downstream tasks such as automated trait identification, growth stage monitoring, and disease detection.

The model will be pretrained using state-of-the-art self-supervised approaches, including masked image modelling and contrastive learning, and enriched with agricultural domain knowledge through carefully designed pretext tasks, synthetic data augmentation, and semantic alignment with textual descriptions of plant traits, phenology, stress symptoms, etc.

A key research challenge will be ensuring generalisation across crop species, imaging conditions, and environmental contexts. AgriFM-G will be validated on a first set of benchmark tasks defined collaboratively by the project's use cases, evaluating accuracy, robustness, and generalisation capacity of the model in isolation before downstream integration with AgriFM-E and AgriFM-M.

Tasks

The main tasks to be performed by the PhD consist of:

1.- Multi-modal dataset preparation:

Identify the potential sources of data relevant to the task, including agriculture-relevant imagery and textual descriptions of agricultural varieties, stress symptoms, etc.

2.- Single and multi-modal pretraining for plant phenotyping:

Design and evaluate self-supervised pretraining strategies that jointly exploit all the available data modalities, including per-modality and with alignment between modalities.

3.- Generalisation across crops and conditions:

Develop fine-tuning and evaluation protocols that allow to assess the generalisation of AgriFM-G to new crop species, growth stages, and acquisition conditions with minimal labelled data.

4.- Benchmark design and evaluation:

Contribute to the definition and evaluation of a benchmark suite for phenotyping FMs, covering trait estimation, growth stage classification, and anomaly/disease detection.

Principales activités

The selected PhD candidate is expected to:

- Become familiar with the state-of-the-art in vision and vision-language Foundation Models.
- Coordinate with other project partners the construction of the datasets.
- Perform large-scale training of vision and vision-language models on an HPC server.
- Write papers describing this work aiming at top computer science venues.

Compétences

- Python programming.
- Deep learning frameworks (preferably Pytorch).
- Use of Linux GPU severs via command line.
- Written scientific English.
- Experience with computer vision, with experience on computer vision for biological applications a plus.
- Experience with language and vision-language models.
- Familiarity with Foundation Models, self-supervised pretraining, or large-scale multi-modal learning.
- Interest in plant sciences, agronomy, or biodiversity.

Avantages

- Subsidized meals
- Partial reimbursement of public transport costs
- Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
- Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
- Professional equipment available (videoconferencing, loan of computer equipment, etc.)
- Social, cultural and sports events and activities
- Access to vocational training
- Social security coverage

Rémunération

Duration: 36 months

Gross Salary per month: 2300€ per month