Untangling Local and Global Deformations in Deep Convolutional Networks for Image Classification and Sliding Window Detection

Abstract : Deep Convolutional Neural Networks (DCNNs) commonly use generic 'max-pooling' (MP) layers to extract deformation-invariant features, but we argue in favor of a more refined treatment. First, we introduce epitomic con-volution as a building block alternative to the common convolution-MP cascade of DCNNs; while having identical complexity to MP, Epitomic Convolution allows for pa-rameter sharing across different filters, resulting in faster convergence and better generalization. Second, we introduce a Multiple Instance Learning approach to explicitly accommodate global translation and scaling when training a DCNN exclusively with class labels. For this we rely on a 'patchwork' data structure that efficiently lays out all image scales and positions as candidates to a DCNN. Factoring global and local deformations allows a DCNN to 'focus its resources' on the treatment of non-rigid defor-mations and yields a substantial classification accuracy improvement. Third, further pursuing this idea, we develop an efficient DCNN sliding window object detector that employs explicit search over position, scale, and aspect ratio. We provide competitive image classification and localization results on the ImageNet dataset and object detection results on the Pascal VOC 2007 benchmark.
Type de document :
Rapport
[Research Report] Toyota Technological Institute at Chicago; Ecole Centrale Paris; Inria Saclay Ile de France. 2014
Liste complète des métadonnées

Littérature citée [38 références]  Voir  Masquer  Télécharger

https://hal.archives-ouvertes.fr/hal-01109289
Contributeur : Stavros Tsogkas <>
Soumis le : lundi 26 janvier 2015 - 04:37:25
Dernière modification le : jeudi 7 février 2019 - 14:00:02
Document(s) archivé(s) le : lundi 27 avril 2015 - 10:10:59

Fichier

PKS_untanglingDCNN_arxiv14.pdf
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-01109289, version 1

Collections

Citation

George Papandreou, Iasonas Kokkinos, Pierre-André Savalle. Untangling Local and Global Deformations in Deep Convolutional Networks for Image Classification and Sliding Window Detection. [Research Report] Toyota Technological Institute at Chicago; Ecole Centrale Paris; Inria Saclay Ile de France. 2014. 〈hal-01109289〉

Partager

Métriques

Consultations de la notice

215

Téléchargements de fichiers

320