2020-02794 - Post-Doctoral Research Visit F/M Detection of abnormal behavior in satellite videos/ Détection de comportements anormaux dans des vidéos satellitaires
Le descriptif de l’offre ci-dessous est en Anglais

Type de contrat : CDD

Niveau de diplôme exigé : Thèse ou équivalent

Fonction : Post-Doctorant

A propos du centre ou de la direction fonctionnelle

The Inria Sophia Antipolis - Méditerranée center counts 34 research teams as well as 8 support departments. The center's staff (about 500 people including 320 Inria employees) is made up of scientists of different nationalities (250 foreigners of 50 nationalities), engineers, technicians and administrative staff. 1/3 of the staff are civil servants, the others are contractual agents. The majority of the center’s research teams are located in Sophia Antipolis and Nice in the Alpes-Maritimes. Four teams are based in Montpellier and two teams are hosted in Bologna in Italy and Athens. The Center is a founding member of Université Côte d'Azur and partner of the I-site MUSE supported by the University of Montpellier.

Contexte et atouts du poste

Weakly-Supervised Anomaly Detection.


Video representation for discriminating anomalies from normal visual patterns is a challenging task in video monitoring domain. The current anomaly detection models lack in providing discriminative video representation for a wide variety of anomalous events. This is due to under-specified problem and unavailability of large-scale videos pertaining to anomalous events, which is a crucial factor for training the modern deep architectures. Thus, we will study the task of anomaly detection which can be leveraged for anomalous video dataset collected from satellites, where videos are split into 2 sets, normal (large set) and anomalous (small set) with only video level labels. The evaluation of proposed framework and model will be performed on an inhouse benchmark dataset which contain human activities of repetitive daily routines in the context of satellite monitoring.


This position is part of the LiChIE project in collaboration with AirBus

Mission confiée

The topic proposed for this position is that of "Weakly-Supervised Anomaly Detection".

In this task, we will study a weakly supervised model performing anomaly detection based on Multiple Instance Learning (MIL) [Mar-97] with ranking loss. By incorporating ranking loss, a model learns to maximize the margin of separation between anomalous and normal instances. In our model, we will incorporate an attention mechanism to focus on the video segments containing the anomalies for learning discriminative representation for the anomalies. The challenge is that these segments are rare and could correspond to only subtle motion/gesture.

[Mar-97] O. Maron and T. Lozano-Perez, “A framework for multiple-instance learning,” in Proceedings of the 1997 Conference on Advances in Neural Information Processing Systems 10, ser. NIPS ’97. Cambridge, MA, USA: MIT Press, 1998, p. 570–576.

Principales activités

The main activity is to develop a novel algorithm for Weakly-Supervised Anomaly Detection

The targeted learning methods are grouped together under the umbrella title of weakly supervised learning, a term indicating learning from incomplete or inexact annotations. Manual annotation in the field of video monitoring is a time-consuming and costly process, which is prone to many challenges such as inter-annotator variance. Compared to bounding box annotations, collecting information about event, is much easier and faster. Research in weakly supervised learning, systems have the capacity to make better and faster utilization of the wealth of image and video data available today. By eliminating the need for exact manually annotated bounding boxes, it removes a basic bottleneck in the machine learning pipeline. Considering the expected widespread applicability of surveillance applications, a weakly supervised learning system makes adjustments to previously trained systems to new environments (a process called fine-tuning in deep learning literature) easier and faster.


  1.  Strong programming skills in Python and in using one of the deep learning libraries like PyTorch or TensorFlow.
  2. Experience in C/C++ and CUDA programming is highly preferred.
  3. Experience with Linux scripting and large scale computations will be an added plus.


  • 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


Gross Salary: 2653 € per month