Less-supervised Segmentation with CNNs: Scenarios, Models and Optimization.

Less-supervised Segmentation with CNNs: Scenarios, Models and Optimization reviews recent progress in deep learning for image segmentation under scenarios with limited supervision, with a focus on medical imaging. It presents the main approaches and state-of-the-art models and presents a broad array of applications in medical image segmentation, including healthcare, oncology, cardiology and neuroimaging. A key objective is to make this mathematical subject accessible to a broad engineering and computing audience by using a large number of intuitive graphical illustrations. The emphasis is on giving conceptual understanding of the methods to foster easier learning. This book is highly suitable for researchers and graduate students in computer vision, machine learning and medical imaging. Gives a good understanding of the different weak-supervision models (i.e., loss functions and priors) and the conceptual connections between them, providing an ability to choose the most appropriate model for a given application scenario Provides knowledge of several possible optimization strategies for each of the examined losses, giving the ability to choose the most appropriate optimizer for a given problem or application scenario Outlines the main strengths and weaknesses of state-of-the-art approaches Gives the tools to understand and use publicly-available code, as well as customize it for specific objectives INDICE: I - IntroductionII - PreliminariesIII - Different levels of supervisionIII.1. Different supervisionsIII.2. Priors.III.2.a Knowledge driven priorsIII.2.b Data driven priorsIV - A unified view V - Semi-supervised learningV.1. Introduction to the setting.V.2. Adversarial learning V.3. Consistency regularizationV.4. Unsupervised representation learningV.5. Self-paced learning.V.6. Mixed-supervision.V.7. Invited chapter.VI - Unsupervised domain adaptationVI.1. Introduction to the setting.VI.2. Adversarial learning.VI.3. Source-free adaptation.VI.4. Domain generalization?VI.5. Invited chapter.VII - Weakly supervised segmentationVII.1. Introduction to the setting.VII.2. From global cues to pixel labelsVII.3. Constrained CNNsVII.3.a Equality constraints.VII.3.b Constrained CNNs: Inequality constraints.VII.4. Class activation maps based methods.VII.5. Invited chapter/sVIII - Few-shot learningVIII.1. Introduction to the setting.VIII.2. Learning to learn.VIII.3. Data augmentation.VIII.4. Simple baselines.VIII.5. Invited chapterIX - Unsupervised segmentationIX.1. Introduction to the setting.IX.2. Auto-encodersIX.3. Use of the gradientIX.4. Leveraging constraintsIX.5. Invited chapterX - Perspectives and future directions

• ISBN: 978-0-323-95674-1
• Editorial: Academic Press
• Encuadernacion: Rústica
• Páginas: 275
• Fecha Publicación: 01/11/2023
• Nº Volúmenes: 1
• Idioma: Inglés