Single Image Depth Prediction with Wavelet Decomposition - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

Single Image Depth Prediction with Wavelet Decomposition

Michaël Ramamonjisoa
  • Fonction : Auteur
Michael Firman
  • Fonction : Auteur
Jamie Watson
  • Fonction : Auteur
Daniyar Turmukhambetov
  • Fonction : Auteur

Résumé

We present a novel method for predicting accurate depths from monocular images with high efficiency. This optimal efficiency is achieved by exploiting wavelet decomposition, which is integrated in a fully differentiable encoder-decoder architecture. We demonstrate that we can reconstruct high-fidelity depth maps by predicting sparse wavelet coefficients. In contrast with previous works, we show that wavelet coefficients can be learned without direct supervision on coefficients. Instead we supervise only the final depth image that is reconstructed through the inverse wavelet transform. We additionally show that wavelet coefficients can be learned in fully self-supervised scenarios, without access to ground-truth depth. Finally, we apply our method to different state-of-the-art monocular depth estimation models, in each case giving similar or better results compared to the original model, while requiring less than half the multiplyadds in the decoder network.
Fichier principal
Vignette du fichier
ramamonjisoa_cvpr21.pdf (14.92 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03482288 , version 1 (15-12-2021)

Identifiants

  • HAL Id : hal-03482288 , version 1

Citer

Michaël Ramamonjisoa, Michael Firman, Jamie Watson, Vincent Lepetit, Daniyar Turmukhambetov. Single Image Depth Prediction with Wavelet Decomposition. Conference on Computer Vision and Pattern Recognition, 2021, Online, United States. ⟨hal-03482288⟩
22 Consultations
11 Téléchargements

Partager

Gmail Facebook X LinkedIn More