1 code implementation • 30 Oct 2023 • Quentin Bouniot, Angélique Loesch, Romaric Audigier, Amaury Habrard
For specialized and dense downstream tasks such as object detection, labeling data requires expertise and can be very expensive, making few-shot and semi-supervised models much more attractive alternatives.
no code implementations • 25 Oct 2023 • Quentin Bouniot, Romaric Audigier, Angélique Loesch, Amaury Habrard
However, for unsupervised pretraining, the popular contrastive learning requires a large batch size and, therefore, a lot of resources.
no code implementations • 27 Oct 2022 • Yassine Naji, Aleksandr Setkov, Angélique Loesch, Michèle Gouiffès, Romaric Audigier
Abnormal event detection in videos is a challenging problem, partly due to the multiplicity of abnormal patterns and the lack of their corresponding annotations.
Ranked #2 on Anomaly Detection on UCSD Ped2
no code implementations • 7 Mar 2022 • Khalil Bergaoui, Yassine Naji, Aleksandr Setkov, Angélique Loesch, Michèle Gouiffès, Romaric Audigier
This paper addresses video anomaly detection problem for videosurveillance.
Ranked #3 on Anomaly Detection on UCSD Peds2
no code implementations • 24 Jan 2022 • Angelique Loesch, Jaonary Rabarisoa, Romaric Audigier
In video surveillance applications, person search is a challenging task consisting in detecting people and extracting features from their silhouette for re-identification (re-ID) purpose.
no code implementations • 24 Jan 2022 • Angelique Loesch, Romaric Audigier
In addition, we propose HPTR, a new bottom-up multi-task method based on transformers as a fast and scalable baseline.
no code implementations • 7 Jan 2022 • Astrid Orcesi, Romaric Audigier, Fritz Poka Toukam, Bertrand Luvison
This can be an issue for HOI detection methods whose complexity depends on the number of people, targets or interactions.
no code implementations • 24 Dec 2021 • Fabian Dubourvieux, Romaric Audigier, Angélique Loesch, Samia Ainouz, Stéphane Canu
(ii) General good practices for Pseudo-Labeling, directly deduced from the interpretation of the proposed theoretical framework, in order to improve the target re-ID performance.
no code implementations • 15 Oct 2021 • Fabian Dubourvieux, Angélique Loesch, Romaric Audigier, Samia Ainouz, Stéphane Canu
However, the effectiveness of these approaches heavily depends on the choice of some hyperparameters (HP) that affect the generation of pseudo-labels by clustering.
1 code implementation • 5 Feb 2021 • Quentin Bouniot, Romaric Audigier, Angélique Loesch
This leads to SAT (Sinkhorn Adversarial Training), a more robust defense against adversarial attacks.
22 code implementations • 17 Nov 2020 • Thomas Defard, Aleksandr Setkov, Angelique Loesch, Romaric Audigier
We present a new framework for Patch Distribution Modeling, PaDiM, to concurrently detect and localize anomalies in images in a one-class learning setting.
Ranked #1 on on MVTecAD
1 code implementation • 5 Oct 2020 • Quentin Bouniot, Ievgen Redko, Romaric Audigier, Angélique Loesch, Amaury Habrard
In this paper, we consider the framework of multi-task representation (MTR) learning where the goal is to use source tasks to learn a representation that reduces the sample complexity of solving a target task.
no code implementations • 28 Sep 2020 • Quentin Bouniot, Ievgen Redko, Romaric Audigier, Angélique Loesch, Amaury Habrard
To the best of our knowledge, this is the first contribution that puts the most recent learning bounds of meta-learning theory into practice for the popular task of few-shot classification.
no code implementations • 20 Sep 2020 • Fabian Dubourvieux, Romaric Audigier, Angelique Loesch, Samia Ainouz, Stephane Canu
A challenge for re-ID is the performance preservation when a model is used on data of interest (target data) which belong to a different domain from the training data domain (source data).
no code implementations • 13 Jan 2020 • Sanaa Chafik, Astrid Orcesi, Romaric Audigier, Bertrand Luvison
In this paper, we introduce a novel human interaction detection approach, based on CALIPSO (Classifying ALl Interacting Pairs in a Single shOt), a classifier of human-object interactions.