Post-Doctoral Research Visit – Characterization of motion anomalies in videos: Application to the detection of AI-generated videos

Type de contrat : CDD Contexte et atouts du poste Funding: Funding for this postdoc position has not yet been secured. The selection of a postdoc candidate is a prerequisite. This is because the funding sources being sought require that the application be submitted by the postdoc fellow. The duration of the postdoc position may range from 12 to 24 months, depending on the funding secured. Mission confiée Subject: A video is more than just a succession of images because it incorporates the fundamental temporal dimension of motion. Motion indeed carries intrinsic information in videos. This characteristic is essential to the analysis. Additionally, motion properties become more explicit over time. Motion is fully represented by the velocity field computed between two images at every time instant of the video. In the long term, the consecutive velocity fields can be viewed as a multivariate time series. The first objective of the postdoc is to create parsimonious time series through learning that can adequately represent motion content and are semantically suitable for generic tasks such as detection, recognition, and classification. More specifically, we will explore anomaly detection in videos, a challenge shared by many applications, albeit in various forms. We will focus on the characterization of motion anomalies based on learned time series. Motion anomalies can manifest themselves in three main ways: deviations from normal behavior or context (e.g., a vehicle driving the wrong way on a highway), sudden divergences (e.g., a vehicle leaving the road, panic in a crowd), or non-natural motion (e.g., presumably in AI-generated videos). This last category will lead us to address the detection of AI-generated videos, whose rapid rise and growing ability to mimic reality raise major societal and economic issues. While the detection of AI-generated still images has been the subject of much research and even challenges, there are still relatively few methods designed specifically for videos. Methods developed for detecting AI-generated images are not effective for videos because they fail to recognize an essential video characteristic: its temporal dimension. This postdoc will build on our recent work, particularly on salient trajectory detection, long-term unsupervised motion segmentation, and automatic detection of AI-generated content. Key words: Motion in image sequences, time series, learning, anomaly characterization, detection of AI-generated videos References - L. Maczyta, P. Bouthemy, and O. Le Meur. Trajectory saliency detection using consistency-oriented latent codes from a recurrent auto-encoder, IEEE Trans. on Circuits and Systems for Video Technology, 32(4):1724 – 1738, April 2022. - E. Meunier and P. Bouthemy. Segmenting the motion components of a video: A long-term unsupervised model, IEEE Transactions on Pattern Analysis and Machine Intelligence, 48(1):500-511, January 2026. - G. Charbel, N. Kindji, E. Fromont, L. M. Rojas-Barahona, and T. Urvoy. Robust detection of synthetic tabular data under schema variability, 40th Annual AAAI Conf. on Artificial Intelligence (AAAI'2026), Singapore, January 2026. Inria, l'institut national de recherche dans les sciences et technologies du numérique, est en appui de l’État pour les stratégies nationales de recherche et d’innovation du numérique en tant qu'Agence de programmes. Inria mène plus de 300 projets de recherche et d’innovation avec ses 3500 scientifiques, ingénieurs et personnels d’appui, en partenariat avec les universités et l’écosystème numérique (entreprises, entrepreneurs, acteurs publics). Ensemble, nous explorons des domaines clés comme l'intelligence artificielle, la cybersécurité, l’informatique quantique, le Cloud, la transformation numérique de la santé, les jumeaux numériques ou encore les technologies numériques pour la défense. Nous construisons des solutions concrètes telles que des logiciels, des startups technologiques, des partenariats avec les entreprises du tissu national et des formations de pointe. Notre objectif : l’impact scientifique, technologique et industriel au service de la souveraineté numérique de la France.