Showing 1–2 of 2 results for author: Daaloul, C
-
DP-Net: Learning Discriminative Parts for image recognition
Authors:
Ronan Sicre,
Hanwei Zhang,
Julien Dejasmin,
Chiheb Daaloul,
Stéphane Ayache,
Thierry Artières
Abstract:
This paper presents Discriminative Part Network (DP-Net), a deep architecture with strong interpretation capabilities, which exploits a pretrained Convolutional Neural Network (CNN) combined with a part-based recognition module. This system learns and detects parts in the images that are discriminative among categories, without the need for fine-tuning the CNN, making it more scalable than other p…
▽ More
This paper presents Discriminative Part Network (DP-Net), a deep architecture with strong interpretation capabilities, which exploits a pretrained Convolutional Neural Network (CNN) combined with a part-based recognition module. This system learns and detects parts in the images that are discriminative among categories, without the need for fine-tuning the CNN, making it more scalable than other part-based models. While part-based approaches naturally offer interpretable representations, we propose explanations at image and category levels and introduce specific constraints on the part learning process to make them more discrimative.
△ Less
Submitted 23 April, 2024;
originally announced April 2024.
-
Sampling From the Wasserstein Barycenter
Authors:
Chiheb Daaloul,
Thibaut Le Gouic,
Jacques Liandrat,
Magali Tournus
Abstract:
This work presents an algorithm to sample from the Wasserstein barycenter of absolutely continuous measures. Our method is based on the gradient flow of the multimarginal formulation of the Wasserstein barycenter, with an additive penalization to account for the marginal constraints. We prove that the minimum of this penalized multimarginal formulation is achieved for a coupling that is close to t…
▽ More
This work presents an algorithm to sample from the Wasserstein barycenter of absolutely continuous measures. Our method is based on the gradient flow of the multimarginal formulation of the Wasserstein barycenter, with an additive penalization to account for the marginal constraints. We prove that the minimum of this penalized multimarginal formulation is achieved for a coupling that is close to the Wasserstein barycenter. The performances of the algorithm are showcased in several settings.
△ Less
Submitted 4 May, 2021;
originally announced May 2021.
