Search

• Empirical observations on convolutional neural networks are presented. • The size of the receptive field did not guarantee improved accuracy. • We observed that dead pixels exist in modern convolutional neural networks. • The effect of dead pixels is discussed with a solution to resolve them. Deep neural networks have been used in various fields, but their internal behavior in how they understand images is not well known. In this study, we discuss two counterintuitive properties of convolutional neural networks (CNNs). First, we evaluated the size of the receptive field of CNNs with their classification accuracy. Previous studies have attempted to increase the size of the receptive field for performance gain. However, we observed that some CNNs with a smaller receptive field can achieve higher classification accuracy. In this regard, we claim that a larger receptive field does not guarantee improved classification accuracy. Second, using the effective receptive field, we examined the contribution of each pixel to the output of CNN. Intuitively, each pixel is expected to equally contribute to the final output, but we found that there exist pixels in a partially dead state with little contribution to the output. We reveal that the reason for dead pixels lies in even stride operations with odd-sized kernels in CNN and propose a kernel padding method to remove the dead pixels. We demonstrated the vulnerability of CNNs with dead pixels when we detect a noise or small box that is on dead pixels. Our findings on dead pixels should be understood and considered in practical applications of CNN. [ABSTRACT FROM AUTHOR]

• We provide empirical tests on various image corruption using vision transformers. • Vision transformers showed performance degradation on contrast-enhanced images. • We proposed PreLayerNorm for the consistent behavior of positional embedding. • We observed that PreLayerNorm improved performance on contrast-enhanced images. • We provide theoretical analyses on the inconsistent behavior of vision transformers. [Display omitted] Vision Transformers (ViTs) have recently demonstrated state-of-the-art performance in various vision tasks, replacing convolutional neural networks (CNNs). However, because ViT has a different architectural design than CNN, it may behave differently. To investigate whether ViT has a different performance or robustness, we tested ViT and CNN under various imaging conditions in practical vision tasks. We confirmed that for most image transformations, ViT's robustness was comparable or even better than that of CNN. However, for contrast enhancement, ViT performed particularly poorly. We show that this is because positional embedding in ViT's patch embedding can work improperly when the color scale changes. We demonstrate that the use of PreLayerNorm, a modified patch embedding structure, ensures the consistent behavior of ViT. Results demonstrate that ViT with PreLayerNorm exhibited improved robustness in the contrast-varying environments. [ABSTRACT FROM AUTHOR]

• A weakly supervised power line detection algorithm is proposed. • A line specialized refinement algorithm is applied for performance improvement. • A repeated learning framework with noise label develops the model output. Detection of power lines in aerial images is an important problem to prevent accidents of unmanned aerial vehicles operating at low altitudes in the electrical industry. Recently, pixel-level power line detection using deep learning has been studied but production of the pixel-level annotations for massive dataset is difficult. In this study, we propose a power line detection algorithm using weakly supervised learning method to reduce the labeling cost for dataset generation. The algorithm is divided into two stages. First, an approximately localized mask was generated based on a convolutional neural network which was trained with only patch-level labels. Second, recursive training of segmentation network with refined broken line segments was executed. A refinement algorithm, line segment connecting (LSC) is a power-line-specialized refinement module that connects broken lines by approximating the segments as partially straight. In proposed algorithm, predicted image at each recursive step was updated as a label of the next training and the label was developed by itself with LSC. The comprehensive experimental results of our algorithm showed state-of-art F 1 -score of 94.3% in weakly supervised learning approaches on public dataset. This result suggests that the proposed algorithm is useful for low labeling cost with high performance in line detection application. [ABSTRACT FROM AUTHOR]