Auswahl der wissenschaftlichen Literatur zum Thema „Interpolation-Based data augmentation“

One of the key problems in the field of increasing the efficiency of neural network tools intended for the analysis of graphic materials is the formation of representative training databases. A promising way to overcome this problem is to increase the size of the training sample by applying the augmentation of training examples due to geometric transformations. However, the modern mathematical apparatus for modifying the geometric parameters of images has shortcomings that can reduce the quality of the obtained images or lead to their insufficient compliance with the tasks. The purpose of the work is the formation of mathematical support, which is used to implement geometric transformations of images during the augmentation of training data of neural network tools. The research methodology is based on the theory of digital processing of signals and images, system analysis and the theory of neural networks and involves the determination of key components of the mathematical support for affine transformations and the definition of the mathematical support for calculating the color of pixels when scaling an image using interpolation methods in various application conditions. As a result of the conducted research, mathematical support was formed, which is used to implement geometric transformations of images when augmenting the training data of neural network tools. The key components of the mathematical support of affine transformations related to the determination of the display dimensions of the modified image and the determination of the color of individual points of the modified image are defined. A mathematical apparatus has been defined that allows you to calculate the dimensions of the display area of the modified image, provided that a rectangular display area is provided. It is shown that in the task of augmentation of training data of neural network tools, it is advisable to perform image scaling using proven non-adaptive interpolation methods: nearest neighbor, bilinear interpolation, bicubic interpolation, Lanzosh interpolation, box filter, triangular filter. A mathematical apparatus for calculating the pixel color of a scaled image using each of the above interpolation methods is defined. The conditions that determine the expediency of using the specified methods in solving practical problems are outlined. It is shown that the method of the nearest neighbor is advisable to use in the case of the need to ensure the simplicity of implementation and high productivity of the augmentation procedure, when the noise of small details and the distortion of the shape of objects containing thin lines can be neglected. The bilinear interpolation method is recommended for use under the conditions of ensuring a balance between the quality of the scaled image and computational costs, and the bicubic interpolation method - under the condition that there are no strict limits on the computational resource intensity of the processing process, when it is necessary to achieve high image quality during scaling. The Lanzosh interpolation method is recommended for scaling images in cases where maximum preservation of quality and detail is required. The box method and the triangular filter method are appropriate to use when the speed of scaling is more important than the quality of the image obtained after scaling. Prospects for further research are related to the development of a methodology for adapting the means of integrated modification of geometry and visual characteristics of images to the conditions of augmentation of training data of neural network systems for video stream analysis.

Action recognition, an essential component of computer vision, plays a pivotal role in multiple applications. Despite significant improvements brought by Convolutional Neural Networks (CNNs), these models suffer performance declines when trained with discontinuous video frames, which is a frequent scenario in real-world settings. This decline primarily results from the loss of temporal continuity, which is crucial for understanding the semantics of human actions. To overcome this issue, we introduce the 4A (Action Animation-based Augmentation Approach) pipeline, which employs a series of sophisticated techniques: starting with 2D human pose estimation from RGB videos, followed by Quaternion-based Graph Convolution Network for joint orientation and trajectory prediction, and Dynamic Skeletal Interpolation for creating smoother, diversified actions using game engine technology. This innovative approach generates realistic animations in varied game environments, viewed from multiple viewpoints. In this way, our method effectively bridges the domain gap between virtual and real-world data. In experimental evaluations, the 4A pipeline achieves comparable or even superior performance to traditional training approaches using real-world data, while requiring only 10% of the original data volume. Additionally, our approach demonstrates enhanced performance on In-the-wild videos, marking a significant advancement in the field of action recognition. The full version of this paper, along with the code and data, can be found at [41].

To solve the problem of poor performance of deep neural network models due to insufficient data, a simple yet effective interpolation-based data augmentation method is proposed: MSMix (Manifold Swap Mixup). This method feeds two different samples to the same deep neural network model, and then randomly select a specific layer and partially replace hidden features at that layer of one of the samples by the counterpart of the other. The mixed hidden features are fed to the model and go through the rest of the network. Two different selection strategies are also proposed to obtain richer hidden r

Data augmentation plays a critical role in improving model performance across various domains, but it becomes challenging with graph data due to their complex and irregular structure. To address this issue, we propose EPIC (Edit Path Interpolation via learnable Cost), a novel interpolation-based method for augmenting graph datasets. To interpolate between two graphs lying in an irregular domain, EPIC leverages the concept of graph edit distance, constructing an edit path that represents the transformation process between two graphs via edit operations. Moreover, our method introduces a context

Compared with traditional sequential learning models, graph-based neural networks exhibit excellent properties when encoding text, such as the capacity of capturing global and local information simultaneously. Especially in the semi-supervised scenario, propagating information along the edge can effectively alleviate the sparsity of labeled data. In this paper, beyond the existing architecture of heterogeneous word-document graphs, for the first time, we investigate how to construct lightweight non-heterogeneous graphs based on different linguistic information to better serve free text represe