Self-supervised pre-raining of Vision Transformers for dense prediction tasks - CEA - Commissariat à l’énergie atomique et aux énergies alternatives Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Self-supervised pre-raining of Vision Transformers for dense prediction tasks

Résumé

We present a new self-supervised pre-training of Vision Transformers for dense prediction tasks. It is based on a contrastive loss across views that compares pixel-level representations to global image representations. This strategy produces better local features suitable for dense prediction tasks as opposed to contrastive pre-training based on global image representation only. Furthermore, our approach does not suffer from a reduced batch size since the number of negative examples needed in the contrastive loss is in the order of the number of local features. We demonstrate the effectiveness of our pre-training strategy on two dense prediction tasks: semantic segmentation and monocular depth estimation.
Fichier principal
Vignette du fichier
vit_dense_pretraining.pdf (1.09 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

cea-04510565 , version 1 (19-03-2024)

Identifiants

  • HAL Id : cea-04510565 , version 1

Citer

Jaonary Rabarisoa, Valentin Belissen, Florian Chabot, Quoc-Cuong Pham. Self-supervised pre-raining of Vision Transformers for dense prediction tasks. T4V @ CVPR 2023 - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) : Transformers for Vision Workshop, Jun 2023, Vancouver, Canada. ⟨cea-04510565⟩
8 Consultations
9 Téléchargements

Partager

Gmail Facebook X LinkedIn More