Zobrazeno 1 - 10
of 740
pro vyhledávání: '"P, Lebailly"'
We identify sufficient conditions to avoid known failure modes, including representation, dimensional, cluster and intracluster collapses, occurring in non-contrastive self-supervised learning. Based on these findings, we propose a principled design
Externí odkaz:
http://arxiv.org/abs/2410.04959
Autor:
Stegmüller, Thomas, Lebailly, Tim, Dukic, Nikola, Bozorgtabar, Behzad, Tuytelaars, Tinne, Thiran, Jean-Philippe
Zero-shot classification capabilities naturally arise in models trained within a vision-language contrastive framework. Despite their classification prowess, these models struggle in dense tasks like zero-shot open-vocabulary segmentation. This defic
Externí odkaz:
http://arxiv.org/abs/2406.16085
Autor:
Lebailly, Tim, Stegmüller, Thomas, Bozorgtabar, Behzad, Thiran, Jean-Philippe, Tuytelaars, Tinne
Leveraging nearest neighbor retrieval for self-supervised representation learning has proven beneficial with object-centric images. However, this approach faces limitations when applied to scene-centric datasets, where multiple objects within an imag
Externí odkaz:
http://arxiv.org/abs/2310.07855
Autor:
Stegmüller, Thomas, Lebailly, Tim, Bozorgtabar, Behzad, Tuytelaars, Tinne, Thiran, Jean-Philippe
Learning dense visual representations without labels is an arduous task and more so from scene-centric data. We propose to tackle this challenging problem by proposing a Cross-view consistency objective with an Online Clustering mechanism (CrOC) to d
Externí odkaz:
http://arxiv.org/abs/2303.13245
Autor:
Lebailly, Tim, Stegmüller, Thomas, Bozorgtabar, Behzad, Thiran, Jean-Philippe, Tuytelaars, Tinne
Most self-supervised methods for representation learning leverage a cross-view consistency objective i.e., they maximize the representation similarity of a given image's augmented views. Recent work NNCLR goes beyond the cross-view paradigm and uses
Externí odkaz:
http://arxiv.org/abs/2303.13606
Autor:
Lebailly, Tim, Tuytelaars, Tinne
The downstream accuracy of self-supervised methods is tightly linked to the proxy task solved during training and the quality of the gradients extracted from it. Richer and more meaningful gradients updates are key to allow self-supervised methods to
Externí odkaz:
http://arxiv.org/abs/2207.14676
Human motion prediction is a necessary component for many applications in robotics and autonomous driving. Recent methods propose using sequence-to-sequence deep learning models to tackle this problem. However, they do not focus on exploiting differe
Externí odkaz:
http://arxiv.org/abs/2010.03006
Autor:
Clément Pruvot, Anne Laure Messagier, Dea Garcia-Hermoso, Fréderic Lebailly, Cédric Aglae, Nicole Desbois-Nogard
Publikováno v:
Medical Mycology Case Reports, Vol 39, Iss , Pp 18-22 (2023)
We report a case of subcutaneous mycosis in the form of a subcutaneous cyst of the index finger, successfully treated by surgery and posaconazole in an 84-year-old female kidney transplant patient. Intra-operative mycological analysis enabled the dia
Externí odkaz:
https://doaj.org/article/29c6ead8b14741f9b23e9f59d244c085
Akademický článek
Tento výsledek nelze pro nepřihlášené uživatele zobrazit.
K zobrazení výsledku je třeba se přihlásit.
K zobrazení výsledku je třeba se přihlásit.
Qualities and Origins of Out-of-Home Food Products: Midday Meals in the Cities of Southeastern Gabon
Publikováno v:
Biology and Life Sciences Forum, Vol 30, Iss 1, p 7 (2023)
The place of local products in out-of-home catering and the growth of non-communicable metabolic diseases in sub-Saharan Africa raise questions. To identify the influence of the economic and social environments on this phenomenon, this study analyzes
Externí odkaz:
https://doaj.org/article/bf6c7ae8c4334c26ae24a3448fa948cd