SFNet: Learning Object-aware Semantic Correspondence - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

SFNet: Learning Object-aware Semantic Correspondence

Résumé

We address the problem of semantic correspondence, that is, establishing a dense flow field between images depicting different instances of the same object or scene category. We propose to use images annotated with binary foreground masks and subjected to synthetic geometric deformations to train a convolutional neural network (CNN) for this task. Using these masks as part of the supervisory signal offers a good compromise between semantic flow methods, where the amount of training data is limited by the cost of manually selecting point correspondences, and semantic alignment ones, where the regression of a single global geometric transformation between images may be sensitive to image-specific details such as background clutter. We propose a new CNN architecture, dubbed SFNet, which implements this idea. It leverages a new and differentiable version of the argmax function for end-to-end training, with a loss that combines mask and flow consistency with smoothness terms. Experimental results demonstrate the effectiveness of our approach, which significantly outperforms the state of the art on standard benchmarks.
Fichier principal
Vignette du fichier
SFNet (1).pdf (3.58 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02088666 , version 1 (03-04-2019)
hal-02088666 , version 2 (05-04-2019)
hal-02088666 , version 3 (21-10-2020)

Identifiants

  • HAL Id : hal-02088666 , version 3

Citer

Junghyup Lee, Dohyung Kim, Jean S Ponce, Bumsub Ham. SFNet: Learning Object-aware Semantic Correspondence. CVPR 2019 - Computer Vision and Pattern Recognition, Jun 2019, Longbeach, United States. ⟨hal-02088666v3⟩
314 Consultations
183 Téléchargements

Partager

Gmail Facebook X LinkedIn More