Skip to main content
Tallo logoTallo logo
Apply for this opportunity

This job application is on an outside website. Be sure to review the job posting there to verify it's the same.

Study on compression of Visual Object Tracking models

Job

ISL - French-German research Institute of Saint-Louis

Saint Louis, MO (In Person)

Full-Time

Posted 3 weeks ago (Updated 2 weeks ago) • Actively hiring

Expires 7/19/2026

Review key factors to help you decide if the role fits your goals.
Pay Growth
?
out of 5
Not enough data
Not enough info to score pay or growth
Job Security
?
out of 5
Not enough data
Calculating job security score...
Total Score
82
out of 100
Average of individual scores

Were these scores useful?

Skill Insights

Compare your current skills to what this opportunity needs—we'll show you what you already have and what could strengthen your application.

Job Description

Study on compression of Visual Object Tracking models ISL - French-German research Institute of Saint-Louis Saint-Louis Speaks German, English, French Apply Summary by Impactpool The internship focuses on the compression of Visual Object Tracking models within the Smart munition division. The candidate will join the IGC group to design and evaluate model compression techniques aimed at reducing model size and complexity while maintaining performance. Responsibilities include conducting literature reviews, implementing and evaluating compression methods, and validating solutions on embedded platforms. The role requires strong programming skills, knowledge of machine learning frameworks, and familiarity with embedded AI hardware.
Candidate Requirements:
Etude sur la compression de modèle de suivi visuel d'objets Mots-clés : edge AI, vision par ordinateur, Vision Transformer, traitement d'image Quelle sera votre mission ? Au sein de la division « Smart munition », vous intégrerez le groupe IGC (Intelligent Guidance and Control), responsable de la conception et du développement d'algorithmes de Visual Object Tracking (VOT) pour le guidage terminal des munitions. Le VOT (cf. Fig. 1) consiste à localiser et suivre automatiquement un objet cible spécifique à travers les images successives d'une séquence vidéo. Selon la littérature récente, les modèles les plus performants reposent principalement sur les Vision Transformers, qui offrent des améliorations significatives par rapport aux réseaux CNN traditionnels. Cependant, ces modèles ne sont pas directement adaptés aux applications embarquées. L'objectif de ce stage est de concevoir et d'évaluer des techniques de compression de modèles afin de réduire la taille et la complexité des modèles, diminuant ainsi le temps d'inférence tout en maintenant les performances de suivi. Vos missions incluront : La réalisation d'une étude bibliographique sur les techniques de compression de modèles L'Implémentation et l'évaluation des différentes méthodes de compression Valider la solution retenue sur une plateforme embarquée lors d'essais en vol Profil recherché Vous êtes actuellement en dernière année de Master ou d'un cursus équivalent, avec une spécialisation dans l'un ou plusieurs des domaines suivants : apprentissage automatique, vision par ordinateur, data science, développement logiciel, mathématiques appliquées ou tout domaine connexe. Compétences et/ou expériences requises : Solides compétences en programmation Python Capacité à effectuer une étude bibliographique et à synthétiser des résultats techniques Bonne maîtrise des frameworks de machine learning (PyTorch) Capacité à implémenter et évaluer des modèles d'apprentissage profond Atouts supplémentaires : Connaissance des plateformes matérielles pour l'IA embarquée (ex : NVIDIA Jetson) Compréhension des techniques de compression de modèles (pruning, quantification, distillation de connaissances…) Expérience avec les outils de gestion de version (ex : Git) Familiarité avec le système d'exploitation Linux Compte tenu de l'environnement international de l'institut, un bon niveau d'anglais est requis. La maîtrise de l'allemand est un plus. Quelles ressources aurez-vous ? Dans le cadre de votre stage, vous bénéficierez d'un ensemble complet de ressources pour soutenir vos activités de recherche, développement et validation, notamment : Un cluster IA haute performance équipé de plusieurs GPU (A100, L40S) pour l'entraînement, l'implémentation et l'évaluation de modèles sur des jeux de données personnalisés Un jeu de données VOT (Visual Object Tracking) sur mesure, avec des scénarios à forte pertinence militaire Des modules Nvidia Jetson pour des tests en laboratoire et l'optimisation logicielle sur matériel embarqué Un drone quadricoptère équipé d'un module NVIDIA Jetson, permettant de valider vos algorithmes en conditions réelles, lors de scénarios de vol Vous serez intégré(e) à une équipe composée d'un chercheur, deux ingénieurs, et un doctorant, tous spécialisés dans les applications de suivi, et collaborerez avec plus de 20 ingénieurs, chercheurs et doctorants au sein du groupe IGC. Annonce publiée le 07/08/2025 Study on compression of Visual Object Tracking models
Keyword:
edge AI, computer vision, transformer, image processing What will be your mission? As part of the "Smart munition" division, you will join the IGC (Intelligent Guidance and Control) group, which is responsible for designing and developing Visual Object Tracking (VOT) algorithms for the terminal guidance of munitions. VOT (see fig. 1) involves automatically locating and tracking a specific target object across consecutive frames in a video sequence. According to recent literature, top-performing models are primarily based on Vision Transformers, which offer significant performance improvements over traditional CNN-based networks. However, these models are not directly suited for embedded applications. The objective of this internship is to design and evaluate model compression techniques to reduce model size and complexity, thereby decreasing inference time while maintaining tracking performance.
Your tasks will include:
Conducting a literature review on model compression techniques Implementing and evaluating various model compression methods Validating the selected solution on an embedded platform during flight tests Who are you? You are currently in your final year of a Master degree or an equivalent program, specializing in one or more of the following areas: Machine learning, Computer vision, Data science, Software development, Applied mathematics or any related field.
Required Competencies and/or Experience on:
Strong programming skills in Python Ability to conduct literature reviews and summarize technical findings Solid understanding of machine learning frameworks (PyTorch) Ability to implement and evaluate deep learning models Additional assets: Familiarity with hardware platforms for embedded AI (e.g., NVIDIA Jetson) Understanding of model compression techniques (such as pruning, quantization, or knowledge distillation) Experience with version control tools (e.g., Git) Familiarity with the Linux operating system Given the institute's international environment, a good level of English is required. Speaking in German or French is a plus. What resources will you have? As part of your internship, you will benefit from a comprehensive set of resources designed to support your research, development, and validation activities.
These include:
High-performance AI cluster equipped with multiple GPUs (A100, L40S) for training, implementing, and evaluating models on custom datasets Custom Visual Object Tracking (VOT) Dataset with military relevant scenarios Nvidia Jetson devices for laboratory-based testing and optimization of your software on embedded hardware A quadcopter UAV equipped with a NVIDIA Jetson module, providing the capability to validate your algorithms in real-world, in-flight scenarios You will be integrated into a team consisting of two engineers, one researcher, and one PhD, all focused on tracking applications, and you will collaborate with more than 20 engineers, scientists, and PhDs within the IGC group. Published on 07/08/2025 At Impactpool we do our best to provide you the most accurate info, but closing dates may be wrong on our site. Please check on the recruiting organization's page for the exact info. Candidates are responsible for complying with deadlines and are encouraged to submit applications well ahead. Before applying, please make sure that you have read the requirements for the position and that you qualify. Applications from non-qualifying applicants will most likely be discarded by the recruiting manager. Summary by Impactpool The internship focuses on the compression of Visual Object Tracking models within the Smart munition division. The candidate will join the IGC group to design and evaluate model compression techniques aimed at reducing model size and complexity while maintaining performance. Responsibilities include conducting literature reviews, implementing and evaluating compression methods, and validating solutions on embedded platforms. The role requires strong programming skills, knowledge of machine learning frameworks, and familiarity with embedded AI hardware.
Candidate Requirements:
Fellowship Master the recruitment process of the impact sector! Become a fellow and gain access to Impactpool's premium material that will boost your career in the impact sector Explore fellowship Show ISL - French-German research Institute of Saint-Louis that you are interested in them Recruiters from the organization will be able to view your profile, contact you for current & future vacancies and engage you on opportunities that match your skills and interest. Connect with organization