Tomato leaf diseases can incur significant financial damage by having adverse impacts on crops and, consequently, they are a major concern for tomato growers all over the world. The diseases may come in a variety of forms, caused by environmental stress and various pathogens. An automated approach to detect leaf disease from images would assist farmers to take effective control measures quickly and affordably. Therefore, the proposed study aims to analyze the effects of transformer-based approaches that aggregate different scales of attention on variants of features for the classification of tomato leaf diseases from image data. Four state-of-the-art transformer-based models, namely, External Attention Transformer (EANet), Multi-Axis Vision Transformer (MaxViT), Compact Convolutional Transformers (CCT), and Pyramid Vision Transformer (PVT), are trained and tested on a multiclass tomato disease dataset. The result analysis showcases that MaxViT comfortably outperforms the other three transformer models with $97\%$ overall accuracy, as opposed to the $89\%$ accuracy achieved by EANet, $91\%$ by CCT, and $93\%$ by PVT. MaxViT also achieves a smoother learning curve compared to the other transformers. Afterwards, we further verified the legitimacy of the results on another relatively smaller dataset. Overall, the exhaustive empirical analysis presented in the paper proves that the MaxViT architecture is the most effective transformer model to classify tomato leaf disease, providing the availability of powerful hardware to incorporate the model.

The field of medical image analysis is in a constant state of evolution, particularly in the challenging tasks of segmenting organs, diseases, and abnormalities. Therefore, in the realm of dental disease diagnosis, image segmentation plays a crucial role in addressing the difficulties faced by dentists worldwide when diagnosing dental diseases with the naked eye. One prominent deep neural network architecture, known as U-Net, originally designed for biomedical image segmentation, has seen multiple variations and advancements aimed at improving its performance. However, the lack of comparative studies has made it challenging to assess the effectiveness of these U-Net variants in segmenting dental X-ray images. The primary objective of this research is to conduct a comprehensive performance comparison among various U-Net architectures for dental image segmentation. Specifically, we examine six U-Net architecture variants: Vanilla U-Net, Dense U-Net, Attention U-Net, SE U-Net, Residual U-Net, and R2 U-Net. These variants employ configurations with two and three convolutional layers in both the encoder and decoder blocks. Our evaluation metrics include Accuracy, Dice coefficient, F1 score, and IoU (Intersection over Union). Among these U-Net variants, Vanilla U-Net, with two convolutional layers, demonstrated the highest level of performance, achieving an Accuracy of 95.56% and an IoU score of 88% on the validation set. Notably, this model also exhibited a shorter processing time compared to the other architectures. Conversely, when employing three convolutional layers, the Dense U-Net variant emerged as the top performer, achieving an Accuracy of 95.94% and an IoU score of 89.07% on the validation set. Furthermore, the segmentation process successfully isolates the teeth from the surrounding structures, which holds promise for improving disease detection through the development of automated disease diagnosis models.

Bio-inspired event cameras have been considered effective alternatives to traditional frame-based cameras for stereo depth estimation, especially in challenging conditions such as low-light or high-speed environments. Recently, deep learning-based supervised event stereo matching methods have achieved significant performance improvements over the traditional event stereo methods. However, the supervised methods depend on ground-truth disparity maps for training, and it is difficult to secure a large amount of ground-truth disparity maps. A feasible alternative is to devise an unsupervised event stereo method that can be trained without ground-truth disparity maps. To this end, we propose the first unsupervised event stereo matching method that can predict dense disparity maps, and is trained by transforming the depth estimation problem into a warping-based reconstruction problem. We propose a novel unsupervised loss function that enforces the network to minimize the feature-level epipolar correlation difference between the ground-truth intensity images and warped images. Moreover, we propose a novel event embedding mechanism that utilizes both temporal and spatial neighboring events to capture spatio-temporal relationships among the events for stereo matching. Experimental results reveal that the proposed method outperforms the baseline unsupervised methods by significant margins (e.g., up to 16.88 % improvement) and achieves comparable results with the existing supervised methods. Extensive ablation studies validate the efficacy of the proposed modules and architectural choices.

Three hundred and sixty degree video is becoming more and more popular on the Internet. By using a Head-Mounted Display, 360-degree video can render a Virtual Reality (VR) environment. However, it is still a big challenge to understand Quality of Experience (QoE) of 360-degree video since user experience during watching 360-degree video is a very complex phenomenon. In this paper, we aim to investigate four QoE aspects of 360-degree video, namely, perceptual quality, presence, cybersickness, and acceptability. In addition, four key QoE-affecting factors of encoding parameters, content motion, rendering device, and rendering mode are considered in our study. To the best of our knowledge, this is the first work that covers a large number of factors and QoE aspects of 360-degree video. In this study, a subjective experiment is conducted using 60 video versions generated from three original 360-degree videos. Based on statistical analysis of the obtained results, various findings on the impacts of the factors on the QoE aspects are provided. In particular, regarding the impacts of encoding parameters, it is found that the difference of QoE is negligible between video versions encoded at 4 K and 2.5 K resolutions. Also, it is suggested that 360-degree video should not be encoded at HD resolution or lower when watching in VR mode using Head Mounted Display. In addition, the bitrate for good QoE varies widely across different video contents. With respect to the content motion factor, its impact is statistically significant on the perceptual quality, presence, and cybersickness. In a comparison of two rendering device sets used in this study, there is no statistically significant difference found for the acceptability and cybersickness. However, the differences of the perceptual quality and presence are indicated to be statistically significant. Regarding the rendering mode, a comparison between VR and non-VR modes is also conducted. Although the non-VR mode always achieves higher perceptual quality scores and higher acceptability rates, more than a half of the viewers prefer the VR mode to the non-VR mode when watching versions encoded at the resolutions of fHD or higher. By contrast, the non-VR mode is preferred at the HD resolution.

Deep-learning-based image inpainting methods have shown significant promise in both rectangular and irregular holes. However, the inpainting of irregular holes presents numerous challenges owing to uncertainties in their shapes and locations. When depending solely on convolutional neural network (CNN) or adversarial supervision, plausible inpainting results cannot be guaranteed because irregular holes need attention-based guidance for retrieving information for content generation. In this paper, we propose two new attention mechanisms, namely a mask pruning-based global attention module and a global and local attention module to obtain global dependency information and the local similarity information among the features for refined results. The proposed method is evaluated using state-of-the-art methods, and the experimental results show that our method outperforms the existing methods in both quantitative and qualitative measures.