MANTRA: Minimum Maximum Latent Structural SVM for Image Classification and Ranking

Thibaut Durand 1 Nicolas Thome 1 Matthieu Cord 1
1 MLIA - Machine Learning and Information Access
LIP6 - Laboratoire d'Informatique de Paris 6
Abstract : In this work, we propose a novel Weakly Supervised Learning (WSL) framework dedicated to learn discriminative part detectors from images annotated with a global label. Our WSL method encompasses three main contributions. Firstly, we introduce a new structured output latent variable model, Minimum mAximum lateNt sTRucturAl SVM (MANTRA), which prediction relies on a pair of latent variables: $h^+$ (resp. $h^-$) provides positive (resp. negative) evidence for a given output $y$. Secondly, we instantiate MANTRA for two different visual recognition tasks: multi-class classification and ranking. For ranking, we propose efficient solutions to exactly solve the inference and the loss-augmented problems. Finally, extensive experiments highlight the relevance of the proposed method: MANTRA outperforms state-of-the art results on five different datasets.
Liste complète des métadonnées

Cited literature [42 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01343784
Contributor : Thibaut Durand <>
Submitted on : Tuesday, July 12, 2016 - 7:57:30 PM
Last modification on : Thursday, March 21, 2019 - 1:05:03 PM

File

mantra_iccv2015.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01343784, version 1

Citation

Thibaut Durand, Nicolas Thome, Matthieu Cord. MANTRA: Minimum Maximum Latent Structural SVM for Image Classification and Ranking. IEEE International Conference on Computer Vision (ICCV15), Dec 2015, Santiago, Chile. ⟨hal-01343784⟩

Share

Metrics

Record views

337

Files downloads

125