- Cet évènement est passé
Séminaire IMAGE: « Deep Image Compression using only Attention-based Neural Networks as analysis and synthesis transforms »
26 octobre 2023 / 14:00 - 15:00
Nous aurons le plaisir d’écouter Natacha Luka, doctorante en 3ème année dans l’équipe IMAGINE du LIGM (ENPC) , à Marne-la-Vallée.
Elle donnera un séminaire IMAGE, le jeudi 26 octobre 2023, à 14h00, en salle de séminaire F-200.
Titre: Deep Image Compression using only Attention-based Neural Networks as analysis and synthesis transforms
Résumé:
In recent years, Learned Image Compression (LIC) has gained prominence for its capacity to compete or outperform traditional handcrafted pipelines, especially at low bit-rates. While existing methods incorporate convolutional priors with occasional attention blocks to address long-range dependencies, recent advances in computer vision advocate for a transformative shift towards fully transformer-based architectures grounded in the attention mechanism. We will investigate the feasibility of image compression exclusively using attention layers in the analysis and synthesis transform. After reviewing the current methods in LIC, we will introduce our proposition based on the concept of learned image queries to aggregate patch information via cross-attention, followed by quantization and coding techniques. Through evaluations, we will discuss our architecture comptiveness compared to convolution counterpart architectures and hand-designed methods across the popular Kodak, DIV2K, and CLIC datasets.