Communication Dans Un Congrès Année : 2024

Diffusion-based image inpainting with internal learning

Résumé

Diffusion models are now the undisputed state-of-the-art for image generation and image restoration. However, they require large amounts of computational power for training and inference. In this paper, we propose lightweight diffusion models for image inpainting that can be trained on a single image, or a few images. We show that our approach competes with large state-of-the-art models in specific cases. We also show that training a model on a single image is particularly relevant for image acquisition modality that differ from the RGB images of standard learning databases. We show results in three different contexts: texture images, line drawing images, and materials BRDF, for which we achieve state-of-the-art results in terms of realism, with a computational load that is greatly reduced compared to concurrent methods.

Fichier non déposé

Dates et versions

hal-05150712 , version 1 (08-07-2025)

Identifiants

Citer

Nicolas Cherel, Andrés Almansa, Yann Gousseau, Alasdair Newson. Diffusion-based image inpainting with internal learning. 2024 32nd European Signal Processing Conference (EUSIPCO), Aug 2024, Lyon, France. pp.446-450, ⟨10.23919/EUSIPCO63174.2024.10714982⟩. ⟨hal-05150712⟩
71 Consultations
0 Téléchargements

Altmetric

Partager

  • More