Seminario del Dpto de Procesamiento de Señales - Pablo Arias - self-supervised learning / image and video restoration

Seminario del Dpto de Procesamiento de Señales - Pablo Arias - self-supervised learning / image and video restoration

de Pablo Muse -
Número de respuestas: 0

Reenvío este mensaje de Lara. Es hoy martes 10/12 a las 14h.

Saludos,

Pablo

-------------

Estimadas y estimados,

Les recuerdo que esta tarde, de 14:00 a 15:00, tendrá lugar el seminario DPS en el salón 725-Beige (piso 7).

Quienes no puedan asistir en persona al seminario, pueden participar conectándose al siguiente enlace: https://salavirtual-udelar.zoom.us/j/86724339649.

La información de la charla se encuentra al final de este correo.

Saludos,

Lara

======================= 


Pablo Arias (Universitat Pompeu Fabra, Barcelona)

Título: Self-supervised training approaches for some multi-frame restoration problems

Abstract: We will present a framework for self-supervised approaches for some multi-frame restoration problems, such as video denoising and burst super-resolution. These approaches allow to learn parameters of regression models -such as neural networks- directly on degraded data, without requiring access to the undegraded ground truth signal. We study the relation between these self-supervised approaches and supervised training for linear degradation operators, and discuss assumptions under which the self-supervised training is equivalent to the supervised one.