Towards Efficient Deep Learning Models For Image Analysis
Download Towards Efficient Deep Learning Models For Image Analysis PDF/ePub or read online books in Mobi eBooks. Click Download or Read Online button to get Towards Efficient Deep Learning Models For Image Analysis book now. This website allows unlimited access to, at the time of writing, more than 1.5 million titles, including hundreds of thousands of titles in various foreign languages.
Towards Efficient Deep Learning Models for Image Analysis
Deep learning has revolutionized computer vision and continues to develop rapidly. As deep learning models are more widely applied to solving various problems, practitioners have imposed more requirements on their efficiency, such as model size, inference speed, accuracy, and adversarial robustness. Generally, more efficient models can be obtained by improving the training method or optimizing deep learning architectures. The first part of this dissertation attempts to modify existing vision transformer (ViT) architectures and design a novel two-phase knowledge distillation framework to obtain smaller, faster, and more accurate ViTs for image classification and regression. Our proposed ViTs also show better transferability in downstream tasks such as object detection. In the second part, we focus on the adversarial robustness of U-Nets, which are popular in medical image segmentation and synthesis. We demonstrate that U-Nets are vulnerable to adversarial attacks, such as the Fast Gradient Sign Method (FGSM), in both tasks. To robustify U-Nets, we not only explore commonly used robust training methods (adversarial training and knowledge distillation), but also propose a neural architecture search method to automatically identify robust architectures. These robust U-Net architectures can achieve high robustness using regular training methods, thus avoiding the sacrifice of accuracy usually brought by robust training methods. Our contributions facilitate the widespread adoption of deep learning models in resource-constrained or security-critical applications.
Computer Vision – ECCV 2024 Workshops
The multi-volume set LNCS 15623 until LNCS 15646 constitutes the proceedings of the workshops that were held in conjunction with the 18th European Conference on Computer Vision, ECCV 2024, which took place in Milan, Italy, during September 29–October 4, 2024. These LNCS volumes contain 574 accepted papers from 53 of the 73 workshops. The list of workshops and distribution of the workshop papers in the LNCS volumes can be found in the preface that is freely accessible online.
Image Analysis and Processing – ICIAP 2022
The proceedings set LNCS 13231, 13232, and 13233 constitutes the refereed proceedings of the 21st International Conference on Image Analysis and Processing, ICIAP 2022, which was held during May 23-27, 2022, in Lecce, Italy, The 168 papers included in the proceedings were carefully reviewed and selected from 307 submissions. They deal with video analysis and understanding; pattern recognition and machine learning; deep learning; multi-view geometry and 3D computer vision; image analysis, detection and recognition; multimedia; biomedical and assistive technology; digital forensics and biometrics; image processing for cultural heritage; robot vision; etc.