A typical semi-supervised learning-based scheme is based on training a single model for labeled data. For unlabeled data, it uses the pseudo-labeling method to obtain labels. However, samples during are often filtered using probability threshold, which suffers from challenge of effective threshold selection. In case high correct may not be labeled, and in low can wrongly labeled. This issue degrades overall performance model. paper addresses this vital by proposing novel approach SSL named...

In prevalent knowledge distillation, logits in most image recognition models are computed by global average pooling, then used to learn encode the high-level and task-relevant knowledge. this work, we solve limitation of logit transfer distillation context. We point out that it prevents informative spatial information, which provides localized as well rich relational information across contexts an input scene. To exploit propose a simple yet effective approach. add local pooling layer branch...

Learning on Grassmann manifold has become popular in many computer vision tasks, with the strong capability to extract discriminative information for imagesets and videos. However, such learning algorithms particularly high-dimensional always involve significantly high computational cost, which seriously limits applicability of more wide areas. In this research, we propose an unsupervised dimensionality reduction algorithm based Locality Preserving Projections (LPP) criterion. LPP is a...

Deep Learning techniques have been successfully used to solve a wide range of computer vision problems. Due their high computation complexity, specialized hardware accelerators are being proposed achieve performance and efficiency for deep learning-based algorithms. However, soft errors, i.e., bit flipping errors in the layer output, often caused due process variation energy particles these systems. These can significantly reduce model accuracy. To remedy this problem, we propose new...

Neural Architecture Search Without Training (NASWOT) has been proposed recently to replace the conventional (NAS). Pioneer works only deploy one or two indicator(s) search. Nevertheless, quantitative assessment for indicators is not fully studied and evaluated. In this paper, we first review several indicators, which are used evaluate network in a training-free manner, including correlation of Jacobian, output sensitivity, number linear regions, condition neural tangent kernel. Our...

Recently, video and image compression methods using neural networks have received much attention. In MPEG standardization, Video Coding for Machine (VCM) is a newly arising topic which attempts to compress features/images the purpose of machine vision tasks. Especially, compressing features has advantages in terms privacy protection computation off-loading. this paper, we propose an effective feature method equipped with super-resolution (SR) module features. Our main motivation comes from...

With the improvements of technologies, software systems have been more complex then ever before. This requires new approaches to improve development processes so that they sufficiently and efficiently meet these challenges. Requirement engineering in this regard facing many problems which conflicts resolution gained very little popularity. In paper we present motivation toward requirement discuss how techniques from other domains can be applied area. We also proposed a approach for based on...

In recent times, the utilization of 3D models has gained traction, owing to capacity for end-to-end training initially offered by Neural Radiance Fields and more recently Gaussian Splatting (3DGS) models. The latter holds a significant advantage inherently easing rapid convergence during offering extensive editability. However, despite advancements, literature still lives in its infancy regarding scalability these this study, we take some initial steps addressing gap, showing an approach...

The performance of an image denoising method highly depends on its ability to produce smoothed homogeneous regions and preserve fine-grained texture details in the restored image. Existing methods rely implicit learning Convolutional Neural Networks (CNNs) restore We show that implicitly learned representations are limited capacity, resulting a sub-optimal pixel estimation loss details. In this work, we argue explicitly introducing low high frequency information enhances representational...

3D models have recently been popularized by the potentiality of end-to-end training offered first Neural Radiance Fields and most Gaussian Splatting models. The latter has big advantage naturally providing fast convergence high editability. However, as research around these is still in its infancy, there a gap literature regarding model's scalability. In this work, we propose an approach enabling both memory computation scalability such More specifically, iterative pruning strategy that...

The versatile nature of Visual Sentiment Analysis (VSA) is one reason for its rising profile. It isn't easy to efficiently manage social media data with visual information since previous research has concentrated on (SA) single modalities, like textual. In addition, most sentiment studies need adequately classify because they are mainly focused simply merging modal attributes without investigating their intricate relationships. This prompted the suggestion developing a fusion deep learning...

Various pruning methods have been proposed to solve the overparameterized problem in deep neural networks. Most of structured used magnitude-based filter importance remove unnecessary filters. Usually, Convolutional Neural Networks (CNN) consist blocks that proceed with batch normalization (BN) operations after convolutional (Conv) operations. Each element is calculated through cooperation between Conv and BN, so both BN parameters must be considered together pruning. However, previous...

Generative Adversarial Networks (GANs) are hindered from real-world applications due to their high computational cost and memory requirements. Model compression techniques, such as quantization, pruning, knowledge distillation, can compress neural networks, lower requirements, model size. However, quantizing generators often leads a suboptimal solution. In this paper, we propose novel method stabilize GAN quantization by the generator discriminator with different bit precision. Our maximizes...