Several recent must see visual reviews are recommended, including Gan, transformer, face super-resolution, remote sensing, etc

Time:2021-8-22

Compile | CV
Report | I love computer vision (wechat ID: aicvml)

This paper introduces the relevant reviews in CV field since the first half of 2021. Package and download are attached at the end of the article.

Gan overview

Gan inverse mapping problem: a comprehensive survey(GAN Inversion: A Survey

Gan inverse mapping refers to transforming a given image into the hidden space of the pre trained Gan model, and the generator can use its inverse mapping code for reliable image reconstruction.

Gan inverse mapping becomes a common space connecting real images and false images, and plays a very important role in image editing tasks of Gan models such as stylegan and biggan. It hides our understanding of Gan hidden space and how to generate passwords with realistic images. Therefore, it is very important to study the inverse mapping of GaN.

Several recent must see visual reviews are recommended, including Gan, transformer, face super-resolution, remote sensing, etc

This review paper focuses on this problem and refers to240 articlesThis paper comprehensively summarizes the algorithms and applications in recent years (important technologies and their applications in image restoration and editing), and points out the future development trends and challenges.

In addition to the paper, the author also established the correspondingGithubWarehouse to further track developments in this area:

https://github.com/weihaox/aw…

Author | Weihao Xia, Yulin Zhang, Yujiu Yang *, Jing Hao Xue, Bolei Zhou *, Ming Hsuan Yang*

Setting: Tsinghua University, Northeastern University, University College London, Chinese University of Hong Kong, University of California Merced

Address| https://arxiv.org/abs/2101.05278

Transformer   overview

The recently launched transformers have “traveled” all directions in the CV field. Later, CV Jun will summarize the application of transformers in the CV field for your reference.

Let’s start with an appetizer  『 Transformers in vision: a survey “, this review aims to provide a comprehensive overview of transformers model in computer vision, which covers the wide application of transformers in the field of computer vision, including popular recognition tasks (such as image classification, target detection, motion recognition and segmentation); Generative models; Multimodal tasks (such as visual question answering and visual reasoning); Video processing (e.g. activity recognition, video prediction); Low level vision (such as image super-resolution and shading); 3D analysis (such as point cloud classification and segmentation). The advantages and limitations of popular technologies are compared from two aspects: architecture design and experimental value.

Finally, the author analyzes the open research direction and possible future work.

Several recent must see visual reviews are recommended, including Gan, transformer, face super-resolution, remote sensing, etc

By Salman Khan, muzammal Naseer, Munawar Hayat, Syed waqas Zamir, Fahad Shahbaz Khan, Mubarak Shah

Unit: mbzuai; Monash University, etc

Thesis| https://arxiv.org/abs/2101.01169

Overview of face super-resolution

Deep Learning-based Face Super-resolution: A Survey

Face super-resolution, also known as face illusion, aims to improve the resolution of one or a sequence of low resolution (LR) face images and generate corresponding high-resolution (HR) face images. It is an image super-resolution problem in a specific field. Recently, face super-resolution has attracted considerable attention and witnessed the dazzling progress of deep learning technology. However, so far, there is little research summary on face super-resolution based on deep learning.

In this survey, the author systematically reviews the deep learning technology in face super-resolution.

Firstly, the problem expression of face super-resolution is summarized.

Secondly, the difference between general image super-resolution and face super-resolution is compared.

Thirdly, the commonly used data sets and performance indicators in face illusion are introduced.

Fourthly, according to the utilization of face specific information, the existing methods are roughly classified. In each category, firstly, the design principles are described, the representative methods are summarized, and the similarities and differences between various methods are compared.

Finally, the prospect of further development of technology in this field is prospected.

Several recent must see visual reviews are recommended, including Gan, transformer, face super-resolution, remote sensing, etc

Author | Junjun Jiang, Chenyang Wang, Xiangming Liu, Jiayi Ma

Setting: Harbin Institute of technology; WuHan University

Thesis| https://arxiv.org/abs/2101.03749

Summary of single classification

One-Class Classification: A Survey

Single class classification (OCC) is a special case of multi class classification. It comes from the data observed by single positive class in the training process. The goal of OCC is to learn a representation and / or a classifier so that it can recognize forward labeled queries in the reasoning process.

In recent years, this topic has attracted considerable attention in the fields of computer vision, machine learning and biometrics. In this investigation, the author investigated the classical statistical methods and the recent visual recognition OCC methods based on deep learning. The advantages and disadvantages of existing OCC methods are discussed, and the promising research direction in this field is determined. In addition, the commonly used data sets and evaluation indexes of OCC are also discussed.

Several recent must see visual reviews are recommended, including Gan, transformer, face super-resolution, remote sensing, etc

author  | Pramuditha Perera, Poojan Oza, Vishal M. Patel

Company  | The Johns Hopkins University

paper  | https://arxiv.org/abs/2101.03064

Overview of personnel re identification

『Deep Learning for Person Re-identification: A Survey and Outlook』

Official account deep learning pedestrian recognition is the latest article in TPAMI 2021. First, the author from Wuhan University, Ye Fan, last week, I also published a detailed interpretation of this article in my computer vision public address. Interested readers can view the review and Prospect of deep learning pedestrian recognition, TPAMI 2021 latest article.

Several recent must see visual reviews are recommended, including Gan, transformer, face super-resolution, remote sensing, etc

Overview of behavior recognition

Human Activity Recognition using Wearable Sensors: Review, Challenges, Evaluation Benchmark

In this survey, the author extensively reviewed the excellent performance methods in human activity recognition based on wearable sensors.

Due to the lack of standardized evaluation, in order to evaluate and ensure the fair comparison between the most advanced technologies, the author uses six public data sets mHealth, uschad, utd-mhad, wisdm, wharf, and opportunity to carry out standardized evaluation benchmarks for the most advanced technologies.

At the same time, an experimental improved method is proposed. Firstly, feature engineering is used to extract features, and then three-layer neural network architecture is used to recognize human activities. Under the same standardized evaluation benchmark, the experiments show that the hybrid experiment has strong generalization ability and high recognition accuracy, and its performance on mHealth, uscad, UTD-1 and utd-2 data sets is better than all advanced technologies.

By | Reem Abdel Salam, Rana Mostafa, mayada hadhood

Setting: Cairo University, Egypt

paper  | https://arxiv.org/abs/2101.01665

Summary of remote sensing land use analysis

Urban land-use analysis using proximate sensing image: a survey is a comprehensive review of the most advanced methods and public data sets of proximate sensing supporting land use analysis.

Author | Zhinan Qiao, Xiaohui yuan

Thesis| https://arxiv.org/abs/2101.04827

A survey of deep neural networks

Hyperbolic Deep Neural Networks: A Survey

This paper makes a coherent and comprehensive review of the literature around the neural components in the construction of hyperbolic deep neural networks and the generalization of leading deep methods in hyperbolic space.

It also introduces the current applications around various machine learning tasks on several publicly available data sets, as well as insightful insights and identifying open problems and promising future directions.

authors | Wei Peng, Tuomas Varanka, Abdelrahman Mostafa, Henglin Shi, Guoying Zhao

Units | University of olu, Finland

paper | https://arxiv.org/abs/2101.04562

Note: the above summary papers can be answered back to the public address in the OpenCV Chinese official account, and get the Baidu cloud download address.

– END –

Compiling: CV Jun

Please contact the official account for authorization.