<html><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><meta http-equiv="content-type" content="text/html; charset=utf-8"><div style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><meta http-equiv="content-type" content="text/html; charset=utf-8"><div style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><meta http-equiv="content-type" content="text/html; charset=utf-8"><div style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><div dir="ltr">Estimados,</div><div dir="ltr"><br></div><div dir="ltr">Es un placer invitarlos a la defensa de la tesis de doctorado en ingeniería eléctrica de Guillermo Carbajal, que tuve el placer de dirigir junto a José Lezama. La defensa tendrá lugar en Facultad de Ingeniería, y será además transmitida por zoom. </div><div dir="ltr"><br></div><div dir="ltr">Van todos los detalles abajo.</div><div dir="ltr"><br></div><div dir="ltr"><div dir="ltr">Saludos,</div><div dir="ltr"><br></div><div dir="ltr">Pablo</div><div dir="ltr"><br></div><div dir="ltr"></div></div><div dir="ltr">—————————</div><div dir="ltr"><br></div><div dir="ltr">Fecha: martes 4 de febrero, 13h30@Uruguay</div><div dir="ltr"><br></div><div dir="ltr">Lugar:</div><div dir="ltr"><br></div><div dir="ltr">- Facultad de Ingeniería, salón 705 (piso 7, cuerpo central). Habrá un brindis a continuación en Instituto de Ingeniería Eléctrica.</div><div dir="ltr"><div class="gmail_quote">
<br>- Zoom : <br><div class="gmail_quote"><a href="https://salavirtual-udelar.zoom.us/j/82021271064?pwd=2CCpF2U4amExffx1kh6g0Mv82AtndD.1">https://salavirtual-udelar.zoom.us/j/82021271064?pwd=2CCpF2U4amExffx1kh6g0Mv82AtndD.1</a></div><div class="gmail_quote"><br></div><div class="gmail_quote">ID de reunión: 820 2127 1064</div><div class="gmail_quote">Código de acceso: mci?0dic3G</div><div class="gmail_quote"><br></div></div><div class="gmail_quote"><br></div><div class="gmail_quote"><br></div><div class="gmail_quote">Título: <b>Single-Image Blind Motion Deblurring: Bridging Blur Formation Models with Data-Driven Learning</b></div><div class="gmail_quote"><br>
Abstract:</div><div class="gmail_quote"><br></div><div class="gmail_quote"><div class="gmail_quote">This thesis addresses the challenging, ill-posed problem of motion deblurring, a prevalent source of image degradation impacting numerous applications, from photography to medical imaging and robotics. Current state-of-the-art deep learning-based deblurring networks, while demonstrating impressive performance on specific datasets, often struggle with generalization to real-world scenarios due to their reliance on learning directly from blurry/sharp image pairs. Conversely, classical model-based approaches, while often generalizing better under their assumed conditions, frequently utilize overly simplified blur models (e.g., uniform or locally uniform blur), limiting their accuracy. </div><div class="gmail_quote"><br></div><div class="gmail_quote">This thesis proposes novel hybrid approaches that synergistically combine the strengths of model-based and data-driven methods. Our first contribution is a motion blur Kernel Prediction Network (KPN), which characterizes non-uniform motion blur using a set of image-adaptive basis kernels and their mixing coefficients. This efficient representation overcomes the limitations of simplistic blur models, enabling accurate modeling of complex, real-world blur patterns. The effectiveness of KPN is demonstrated in diverse datasets and settings. Specifically, the KPN is integrated into two distinct blind deblurring algorithms: one employing an adaptation of the Richardson-Lucy deconvolution algorithm and another leveraging a recent non-blind restoration network. The latter leads to the Joint Motion Kernel Prediction and Deblurring (J-MKPD) blind restoration method, a joint training approach that combines the motion blur KPN with an unrolled plug-and-play restoration network. </div><div class="gmail_quote"><br></div><div class="gmail_quote">Motivated by the high-quality results obtained with J-MKPD, this thesis further explores alternative degradation models that are better suited for specific scenarios. A camera trajectory prediction network is developed to estimate the motion blur kernel field resulting from camera shake. This network, when jointly trained with the same unrolled plug-and-play restoration network as J-MKPD, forms the Joint Motion Trajectory Prediction and Deblurring (J-MTPD) method, specifically designed for camera shake blur. </div><div class="gmail_quote"><br></div><div class="gmail_quote">A third method, Joint Motion Offsets Prediction and Deblurring (J-MOPD), utilizes a more expressive offset-based representation of the motion blur kernel field. While more challenging to train, this representation offers increased flexibility. A novel training strategy is introduced to constrain the solution space, enabling efficient integration with the unrolled plug-and-play network and resulting in a blind deblurring method assuming locally uniform blur. </div><div class="gmail_quote"><br></div><div class="gmail_quote">Recognizing the crucial role of training data in the generalization capabilities of end-to-end deblurring networks, this thesis comprehensively analyzes existing datasets, identifying their limitations and the underlying causes of poor generalization. Based on this analysis, a novel procedural methodology for generating synthetic training data is then proposed. This methodology focuses on capturing the essential characteristics of real-world blur, producing a virtually unlimited supply of diverse, high-quality training pairs. </div><div class="gmail_quote"><br></div><div class="gmail_quote">The resulting J-MKPD, J-MTPD, and J-MOPD methods constitute a suite of publicly available deblurring algorithms demonstrating excellent cross-dataset performance, especially on real-world image datasets. J-MKPD proves effective for motion blur up to 33 pixels, while J-MTPD excels in handling camera shake blur, and J-MOPD is well suited for locally uniform blur. Furthermore, the thesis explores the application of J-MKPD for super-resolution and demonstrates the video generation capabilities of J-MTPD and J-MOPD. This work bridges the gap between model-based and data-driven approaches, providing robust solutions while offering valuable insights into the limitations of existing techniques and the challenges in creating truly generalizable deblurring methods.</div></div><div class="gmail_quote"><div style="orphans: 2; widows: 2;"><font face="Slack-Lato, Slack-Fractions, appleLogo, sans-serif"><br></font></div><div style="orphans: 2; widows: 2;"><font face="Slack-Lato, Slack-Fractions, appleLogo, sans-serif"><br></font></div>
------------------------------------<br>
<br>
Tribunal:</div><div class="gmail_quote"><br></div><div class="gmail_quote">Pablo Arias (revisor externo), Universitat Pompeu Fabra</div><div class="gmail_quote">Matías Di Martino, Universidad Católica del Uruguay</div><div class="gmail_quote">Alicia Fernández, Universidad de la República<br>Matías Tassano (revisor externo), Meta Reality Labs</div><div class="gmail_quote"><br></div><div class="gmail_quote"><br>
Directores de tesis:</div><div class="gmail_quote"><br>Pablo Musé, Universidad de la República</div><div class="gmail_quote">José Lezama, Universidad de la República</div><div class="gmail_quote">
<br>
————————————————————</div><div class="gmail_quote"><br></div></div><br><div><br></div></div></div><div>
----------------------------------------------<br>Pablo Musé, PhD<br>Full Professor of Signal Processing<br>Universidad de la República <br>Uruguay <br>+598 27110974<br>iie.fing.edu.uy/~pmuse/
</div>
<br></div></body></html>