Pré-Publication, Document De Travail Année : 2026

Compressing image encoders via latent distillation

Résumé

Deep learning models for image compression often face practical limitations in hardware-constrained applications. Although these models achieve high-quality reconstructions, they are typically complex, heavyweight, and require substantial training data and computational resources. We propose a methodology to partially compress these networks by reducing the size of their encoders. Our approach uses a simplified knowledge distillation strategy to approximate the latent space of the original models with less data and shorter training, yielding lightweight encoders from heavyweight ones. We evaluate the resulting lightweight encoders across two different architectures on the image compression task. Experiments show that our method preserves reconstruction quality and statistical fidelity better than training lightweight encoders with the original loss, making it practical for resource-limited environments.

Fichier principal
Vignette du fichier
main.pdf (4.17 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05448409 , version 1 (08-01-2026)
hal-05448409 , version 2 (18-05-2026)

Licence

Identifiants

Citer

Caroline Mazini Rodrigues, Nicolas Keriven, Thomas Maugey. Compressing image encoders via latent distillation. 2026. ⟨hal-05448409v1⟩
194 Consultations
164 Téléchargements

Altmetric

Partager

  • More