Caltech Dataset Research Articles

Adaptive class token knowledge distillation for efficient vision transformer

The Vision Transformer (ViT) outperforms Convolutional Neural Networks (CNNs) but at the cost of significantly higher computational demands. Knowledge Distillation (KD) has shown promise in compressing complex networks by transferring knowledge from a large pre-trained model to a smaller one. However, current KD methods for ViT often rely on CNNs as teachers or neglect the importance of class token ([CLS]) information, resulting in ineffective distillation of ViT’s unique knowledge. In this paper, we propose Adaptive Class token Knowledge Distillation ([CLS]-KD), which fully exploits information from the class token and patches in ViT. For class embedding (CLS) distillation, the intermediate CLS of the student model is aligned with the corresponding CLS of the teacher model through a projector. Furthermore, we introduce CLS-patch attention map distillation, where an attention map between the CLS and patch embeddings is generated and matched at each layer. This empowers the student model to learn how to dynamically extract patch embedding information into the CLS under teacher guidance. Finally, we propose Adaptive Layer-wise Distillation (ALD) to mitigate the imbalance in distillation effects varying with the depth of layers. This method assigns greater weight to the losses in layers where the training discrepancies between the teacher and student models are larger during distillation. Through these strategies, [CLS]-KD consistently surpasses existing state-of-the-art methods on the ImageNet-1K dataset across various teacher-student configurations. Furthermore, the proposed method demonstrates its generalization capability through transfer learning experiments on the CIFAR-10, CIFAR-100, and CALTECH-256 datasets.

SNN using color-opponent and attention mechanisms for object recognition

The current spiking neural network (SNN) relies on spike-timing-dependent plasticity (STDP) primarily for shape learning in object recognition tasks, overlooking the equally critical aspect of color information. To address this gap, our study introduces an unsupervised variant of STDP that incorporates principles from color-opponency mechanisms (COM) and classical receptive fields (CRF) found in the biological visual system, facilitating the integration of color information during parameter updates within the SNN architecture. Our approach initially preprocesses images into two distinct feature maps: one for shape and another for color. Then, signals derived from COM and intensity concurrently drive the STDP process, thereby updating parameters associated with both color and shape feature maps. Furthermore, we propose a channel-wise attention mechanism to enhance differentiation among objects sharing similar shapes or colors. Specifically, this mechanism utilizes convolution to generate an output spike-wave, identifying a winner based on earliest spike timing and maximal potential. The winning kernel computes attention, which is then applied via convolution to each input image feature map, generating post-feature maps. A STDP-like normalization rule compares firing times between pre- and post-feature maps, dynamically adjusting channel weights to optimize object recognition during the training phase.We assessed the proposed algorithm using SNN with both single-layer and multi-layer architectures across three datasets. Experimental findings highlight its efficacy and superiority in complex object recognition tasks compared to state-of-the-art (SOTA) algorithms. Notably, our approach achieved a significant 20% performance improvement over the SOTA on the Caltech-101 dataset. Moreover, the algorithm is well-suited for hardware implementation and energy efficiency, leveraging a winner-selection mechanism based on the earliest spike time.

A Mighty Image Retrieval Descriptor Based on Machine Learning and Gaussian Derivative Filter

The development of new image descriptor has always been an important topic to improve the efficiency of content- based image classification and retrieval. Improvements and developments in machine learning and deep learning algorithms as well as artificial intelligence algorithms are widely used by researchers to obtain effective CBIR descriptors. In our article, we will present a robust image descriptor, extended by machine learning and deep learning algorithms. The descriptor is provided through a Gaussian derivative filter scaffold named GDF-HOG with an enhanced convolutional neural network (CNN) AlexNet, to reduce the dimensions we used the principal component analysis algorithm. The experimental results were carried out on Oliva and Torralba, Caltech-101, Wang and Coil100 datasets. Experiments show that the accuracy of the proposed method is 98.23% for Coil-100%, 95.92% for Corel-1000, value 87.17 and 94.6% for Oliva and Torralba. In comparison our results with other descriptor image classifiers show that they achieved accuracy increases of 0.12% on average and up to 3.23%. These experimental results affirm the advantage of the proposed descriptor over existing systems based in terms of average accuracy. the proposed descriptor improves the precision, and also reduces the complexity of the calculation.

Survey of image classification models for transfer learning

Training models for image classification is a very time-consuming task. It has always been a challenge for researchers and practitioners to train a model partly because of the large dataset required, which is complex and sometimes almost impossible to source. This has recently led to the use of pre-trained models for image classifications. Pre-trained models have gained popularity because they initialize the model with appropriate weight and significantly reduce the training time and dataset required. There are many image classification pre-trained models in use today, and this paper will focus on investigating the performance of the ten top models (ConvNext, DenseNet, EfficientNet, InceptionResNet, Inception, mobileNet, ResNet, VGG, Xception, NasNet) using caltech101 dataset containing 101 object classes and caltech256 dataset containing 256 object classes. The models are all trained on the ImageNet-1k dataset. Tensorflow and Keras were used as the frameworks for developing the experiments. The accuracy, precision, recall, and f1-score were used as metrics for model performance evaluation. The findings and analysis underscore the significance of training time, number of epochs, and choice of model in image classification.

Content-based image retrieval using multi-scale averaging local binary patterns

Feature extraction has a significant impact on the accuracy of Content-Based Image Retrieval (CBIR) algorithms since the content of images is encoded in feature vectors. In this paper, an effective method for texture feature extraction is proposed based on local patterns. In the proposed method, first the image is formed in different scales, then the texture features are extracted from the scales of the image. Finally, extracted features of different scales are concatenated to construct the final feature vector. To evaluate the proposed method, five datasets including Corel-1k, Brodatz, VisTex, Corel-10k, STex, Caltech256, and Oliva are used. In the evaluation process, the effect of different scales and different coding schemes on retrieval precision is investigated. The proposed method is compared with existing CBIR models based on local patterns. The proposed method achieves the best precision of 64.16%, 81.66%, 88.59%, 30.63%, and 69.63%, 13.59%, and 64.10% on the Corel-1k, Brodatz, VisTex, Corel-10k, STex, Caltech256, and Oliva datasets, respectively.

AdvRain: Adversarial Raindrops to Attack Camera-Based Smart Vision Systems

Vision-based perception modules are increasingly deployed in many applications, especially autonomous vehicles and intelligent robots. These modules are being used to acquire information about the surroundings and identify obstacles. Hence, accurate detection and classification are essential to reach appropriate decisions and take appropriate and safe actions at all times. Current studies have demonstrated that “printed adversarial attacks”, known as physical adversarial attacks, can successfully mislead perception models such as object detectors and image classifiers. However, most of these physical attacks are based on noticeable and eye-catching patterns for generated perturbations making them identifiable/detectable by the human eye, in-field tests, or in test drives. In this paper, we propose a camera-based inconspicuous adversarial attack (AdvRain) capable of fooling camera-based perception systems over all objects of the same class. Unlike mask-based FakeWeather attacks that require access to the underlying computing hardware or image memory, our attack is based on emulating the effects of a natural weather condition (i.e., Raindrops) that can be printed on a translucent sticker, which is externally placed over the lens of a camera whenever an adversary plans to trigger an attack. Note, such perturbations are still inconspicuous in real-world deployments and their presence goes unnoticed due to their association with a natural phenomenon. To accomplish this, we develop an iterative process based on performing a random search aiming to identify critical positions to make sure that the performed transformation is adversarial for a target classifier. Our transformation is based on blurring predefined parts of the captured image corresponding to the areas covered by the raindrop. We achieve a drop in average model accuracy of more than 45% and 40% on VGG19 for ImageNet dataset and Resnet34 for Caltech-101 dataset, respectively, using only 20 raindrops.

SpinalNet: Deep Neural Network With Gradual Input

Deep neural networks (DNNs) have achieved the state of the art performance in numerous fields. However, DNNs need high computation times, and people always expect better performance in a lower computation. Therefore, we study the human somatosensory system and design a neural network (SpinalNet) to achieve higher accuracy with fewer computations. Hidden layers in traditional NNs receive inputs in the previous layer, apply activation function, and then transfer the outcomes to the next layer. In the proposed SpinalNet, each layer is split into three splits: 1) input split, 2) intermediate split, and 3) output split. Input split of each layer receives a part of the inputs. The intermediate split of each layer receives outputs of the intermediate split of the previous layer and outputs of the input split of the current layer. The number of incoming weights becomes significantly lower than traditional DNNs. The SpinalNet can also be used as the fully connected or classification layer of DNN and supports both traditional learning and transfer learning. We observe significant error reductions with lower computational costs in most of the DNNs. Traditional learning on the VGG-5 network with SpinalNet classification layers provided the state-of-the-art (SOTA) performance on QMNIST, Kuzushiji-MNIST, and EMNIST (Letters, Digits, and Balanced) datasets. Traditional learning with ImageNet pre-trained initial weights and SpinalNet classification layers provided the SOTA performance on STL-10, Fruits 360, Bird225, and Caltech-101 datasets. The scripts of the proposed SpinalNet training are available at the following link: <uri>https://github.com/dipuk0506/SpinalNet</uri> <i>Impact Statement</i>&#x2014;Research in deep neural networks (DNNs) has gained significant attention from industries and academia due to their eye-catching performance. DNNs have enabled machines to perform myriad tasks with high accuracy that once only humans could do. Several researchers have recently proposed different types of NNs and have achieved high accuracy. The recent success of biologically inspired convolutional neural networks and the miraculous spinal architecture of humans has motivated us to develop a neural network with gradual inputs. We have achieved superior performance in several datasets. After the first online appearance of the initial version of this paper, several researchers have applied the proposed neural network in several new datasets and reported promising results. We may observe numerous novel applications of SpinalNet in upcoming years.

The Weights Reset Technique for Deep Neural Networks Implicit Regularization

We present a new regularization method called Weights Reset, which includes periodically resetting a random portion of layer weights during the training process using predefined probability distributions. This technique was applied and tested on several popular classification datasets, Caltech-101, CIFAR-100 and Imagenette. We compare these results with other traditional regularization methods. The subsequent test results demonstrate that the Weights Reset method is competitive, achieving the best performance on Imagenette dataset and the challenging and unbalanced Caltech-101 dataset. This method also has sufficient potential to prevent vanishing and exploding gradients. However, this analysis is of a brief nature. Further comprehensive studies are needed in order to gain a deep understanding of the computing potential and limitations of the Weights Reset method. The observed results show that the Weights Reset method can be estimated as an effective extension of the traditional regularization methods and can help to improve model performance and generalization.

Privileged information learning with weak labels

Privileged information learning is proposed to construct the classifier by incorporating privileged knowledge. At present, most of the privileged information learning methods assume that the instance is accurately labeled. However, in real-world applications, an instance may be weakly labeled. In this paper, we propose a novel privileged information learning method with weak labels (PLWB). The hypothesis of our work is that an instance may be annotated by a number of labelers and different labelers may give different labels to this instance due to distinct professional knowledge and subjective factors. It leads to ambiguous labels of instances, namely weak labels. To solve this problem, our methodology is to give each labeler a weight and incorporate these weights into a privileged information learning model. Our technique is to employ a heuristic framework to optimize the labeler weights and the privileged information learning model jointly. The existing privileged information learning methods do not consider the weak label problem, and assign an equal or random weight to each labeler. Our work is different from these methods. The novelty and theoretical contribution is that this is the first work to deal with the weak label problem in privileged information learning. The merit is that we assign an unknown weight to each labeler and solve the optimal values of these weights in the optimization process, such that the performance of the learning model can be improved with the optimal labeler weights. In the experiments, the tool that we use is MATLAB, in which we implement our algorithm. The experimental datasets include one handwritten categorization dataset, two image classification datasets (i.e., Animals-with-Attributes dataset and Caltech-101 dataset), and one disease diagnosis dataset (i.e., Alzheimer’s Disease Neuroimaging Initiative dataset), in which the number of instances used is 2000, 6180, 8677 and 202, respectively. The obtained results are that: (1) by optimizing the labeler weights, the proposed PLWB method obtains explicitly higher classification accuracy than the existing privileged information learning methods; (2) PLWB has relatively higher training time since it needs to solve the labeler weights in the optimization process.

Evolving Deep Multiple Kernel Learning Networks Through Genetic Algorithms

Today's Industrial Internet of Things (IIoT) have achieved excellent manufacturing efficiency and automation results by leveraging machine learning (ML) and deep learning (DL). However, trustworthiness of ML/DL brings significant challenges to IIoT. This article proposes an evolving deep multiple kernel learning network through genetic algorithm (KNGA). Our KNGA method uses genetic algorithm (GA) to find the best deep multiple kernel learning structure, including the weights and the topology of the model. Compared with the current well-known models, KNGA has advantages in three aspects: 1) It can achieve good results without using many samples during model training; 2) the model can evolve in the process of training, including self-growth, and self-pruning; and 3) its trustworthiness and reliability can be guaranteed. Moreover, the whole model ensures excellent performance and requires manual adjustment of only a few parameters. Extensive experiments on the UCI, KEEL, Caltech256, and MNIST datasets demonstrate the effectiveness and trustworthiness of the proposed method.

End-to-End Object Detection with Enhanced Positive Sample Filter

Discarding Non-Maximum Suppression (NMS) post-processing and realizing fully end-to-end object detection is a recent research focus. Previous works have proved that the one-to-one label assignment strategy provides the chance to eliminate NMS during inference. However, this strategy might also result in multiple predictions with high scores due to the inconsistency of label assignment during training. Thus, how to adaptively identify only one positive sample as a final prediction for each Ground-Truth instance remains important. In this paper, we propose an Enhanced Positive Sample Filter (EPSF) to filter out the single positive sample for each Ground-Truth instance and lower the confidence of other negative samples. This is mainly achieved with two components: a Dual-stream Feature Enhancement module (DsFE) and a Disentangled Max Pooling Filter (DeMF). DsFE makes full use of representations trained with different targets so as to provide rich information clues for positive sample selection, while DeMF enhances the feature discriminability in potential foreground regions with disentangled pooling. With the proposed methods, our end-to-end detector achieves a better performances against existing NMS-free object detectors on COCO, PASCAL VOC, CrowdHuman and Caltech datasets.

An Image Classification Method Based on Adaptive Attention Mechanism and Feature Extraction Network.

The convolution neural network (CNN) not only has high fault tolerance but also has high computing capacity. The image classification performance of CNN has an important relationship with its network depth. The network depth is deeper, and the fitting ability of CNN is stronger. However, a further increase in the depth of CNN will not improve the accuracy of the network but will produce higher training errors, which will reduce the image classification performance of CNN. In order to solve the above problems, this paper proposes a feature extraction network, AA-ResNet with an adaptive attention mechanism. The residual module of the adaptive attention mechanism is embedded for image classification. It consists of a feature extraction network guided by the pattern, a generator trained in advance, and a complementary network. The feature extraction network guided by the pattern is used to extract different levels of features to describe different aspects of an image. The design of the model effectively uses the image information of the whole level and the local level, and the feature representation ability is enhanced. The whole model is trained as a loss function, which is about a multitask problem and has a specially designed classification, which helps to reduce overfitting and make the model focus on easily confused categories. The experimental results show that the method in this paper performs well in image classification for the relatively simple Cifar-10 dataset, the moderately difficult Caltech-101 dataset, and the Caltech-256 dataset with large differences in object size and location. The fitting speed and accuracy are high.

Deep Domain Adaptation Using Cascaded Learning Networks and Metric Learning

How can we bridge efficiently distribution difference between the source and target domains in an isomorphic latent feature space by using metric learning. This study introduces metric learning on manifolds which combine a cascaded learning network and a metric learning model to form a Unified Domain Adaptation Model. Our approach is based on formulating a transfer from the source to the target as a geometric mean metric learning problem on manifolds. The solution of symmetric positive definite covariance matrices not only reduces the statistical differences between the source and target domains, but also the underlying geometry of the source and target domains using diffusions on the underlying source and target manifolds. By retaining both the nonlinear structure of the Riemannian geometry of the open cone of symmetric positive definite matrices and cascaded learning networks, we improve the state-of-the-art results on the Amazon (A), Caltech256 (C), DSLR (D), Webcam (W) and VisDA benchmark datasets by knowledge transfer, while achieving comparable performances to competing methods on domain adaptation modeling.

Curiosity-Driven Class-Incremental Learning via Adaptive Sample Selection

Modern artificial intelligence systems require class-incremental learning while suffering from catastrophic forgetting in many real-world applications. Due to the missing knowledge of past data, performance substantially degrades. Recent methods often used knowledge distillation and bias correction to avoid catastrophic forgetting caused by cognitive bias. However, since these methods mainly learn all samples indiscriminately, the model is hard to learn what it truly needs from the data stream to balance the new and old knowledge, leading to inevitable forgetting. Instead of considering each sample indiscriminately, the model should learn from its curious samples automatically. To tackle this problem, we propose a curiosity-driven class-incremental learning approach via adaptive sample selection for learning a more generalized model with fewer ineffective updates. Specifically, our method quantifies the model’s curiosity in each sample by two properties: uncertainty and novelty. Our model learns informative samples selectively during training utilizing the proposed uncertainty property, which benefits the classification decision boundary. In light of the imbalanced data, a novelty property is used to selectively optimize the model by employing dissimilar samples, endowing it with more robustness and less cognitive bias. Our method successfully reduces catastrophic forgetting and can be flexibly incorporated with other techniques. Extensive experiments and in-depth analysis on the CIFAR-100, Tiny-ImageNet and Caltech-101 datasets show that our approach outperforms competing methods for class-incremental learning in terms of preventing catastrophic forgetting.

An Improved Probability Density Function (PDF) for Face Skin Detection

Face Detection by skin color in the field of computer vision is a difficult challenge. Detection of human skin focuses on the identification of pixels and skin-colored areas of a given picture. Since skin colors are invariant in orientation and size and rapid to process, they are used in the identification of human skin. In addition features like ethnicity, sensor, optics and lighting conditions that are different are sensitive factors for the relationship between surface colors and lighting (an issue that is strongly related to color stability). This paper presents a new technique for face detection based on human skin. Three methods of Probability Density Function (PDF) were applied to detect the face by skin color; these are the Extreme Value Distribution Function and the Exponential Distribution Function methods, in addition to a new proposed model, over the HSV (Hue, Saturation, and Value) color space. The suggested technique aims to enhance skin pixel detection and improve the detection accuracy of a colored region in the human skin in a specific photo. The new model has proved to be 96.05% more accurate than the Extreme value distribution function and Exponential distribution function according to the selected region of the face during experiments. The images used in this paper were 380 color images from CalTech (California Technology Institute) dataset.

Graph Convolutional Network Combined with Semantic Feature Guidance for Deep Clustering

The performances of semisupervised clustering for unlabeled data are often superior to those of unsupervised learning, which indicates that semantic information attached to clusters can significantly improve feature representation capability. In a graph convolutional network (GCN), each node contains information about itself and its neighbors that is beneficial to common and unique features among samples. Combining these findings, we propose a deep clustering method based on GCN and semantic feature guidance (GFDC) in which a deep convolutional network is used as a feature generator, and a GCN with a softmax layer performs clustering assignment. First, the diversity and amount of input information are enhanced to generate highly useful representations for downstream tasks. Subsequently, the topological graph is constructed to express the spatial relationship of features. For a pair of datasets, feature correspondence constraints are used to regularize clustering loss, and clustering outputs are iteratively optimized. Three external evaluation indicators, i.e., clustering accuracy, normalized mutual information, and the adjusted Rand index, and an internal indicator, i.e., the Davidson-Bouldin index (DBI), are employed to evaluate clustering performances. Experimental results on eight public datasets show that the GFDC algorithm is significantly better than the majority of competitive clustering methods, i.e., its clustering accuracy is 20% higher than the best clustering method on the United States Postal Service dataset. The GFDC algorithm also has the highest accuracy on the smaller Amazon and Caltech datasets. Moreover, DBI indicates the dispersion of cluster distribution and compactness within the cluster.

Image classification method based on local receptive field extended S-Mobilenet model

Aiming at the problem that the lightweight deep neural network mobilenet will reduce the classification accuracy, the hollow convolution kernel is introduced into a convolution layer of mobilenet model, and a S-Mobilenet model based on local receptive field expansion is proposed. The model is divided into three structures according to the location of the hole convolution kernel. Without increasing the number of parameters, it can expand the local receptive field of the convolution kernel and improve the classification accuracy. The experiment was carried out on caltech-101 data set, caltech-256 data set and animal classification database of Tubingen University. The results show that S-Mobilenet model can obtain better classification accuracy than Mobilenet, which can be improved by 2% at most.

Fast and effective pedestrian detection based on low-level visual features combination

Automatic pedestrian detection is a crucial task for intelligent vehicles as it helps avoid car-to-pedestrian accidents. Recently, several approaches have been proposed to improve detection accuracy, such as the channel-based family, which is based on the aggregation of channel features (e.g., HOG + LUV) to represent the shape of pedestrians. Previous handcrafted-based detectors significantly improved the detection performance by applying different filters to channel features. However, when there are similar-looking background objects, their detection performance suffers due to the lack of texture features. Another drawback of these approaches lies in their high computational cost, limiting their use for real-world applications, especially for resource constrained systems. To mitigate these drawbacks while maintaining a distinguishable detection accuracy, we propose a method that combines a simple but effective texture descriptor based on local binary patterns with HOG + LUV. The concatenated features are used to train a boosted decision trees classifier. To evaluate our approach, we carried out extensive experiments on four well-known datasets, including INRIA, ETH, KITTI, and Caltech. Experimental results show that the proposed method achieves a competitive performance to the channel-based detectors with significantly lower computational cost; it runs at 91.64 and 6.08 times faster than the closest competitors detectors on the INRIA and Caltech datasets, respectively. Furthermore, it outperforms these methods when evaluating under the corrected annotations of the Caltech dataset.

Methodology for image retrieval based on binary spacepartitioning and perceptual image hashing

The paperfocuses on the content-based image retrieval systems building. The main challengesin the construction of such sys-tems are considered, the components of such systems are reviewed, and a brief overview of the main methods and techniques that have been used in this area to implement the main components of image search systems is given.As one of the options for solving such a problem, an image retrievemethodology based on the binary space partitioning method and the perceptual hashing method is proposed. Space binary partition trees are a data structuresobtained as follows: the space is partitioned by a hyperplane into two half-spaces, and theneach half-space is recursively partitioned until each node contains only a trivial part of the input features. Perceptual hashing algorithms make it possible to represent an image as a 64-bit hash value, with similar images represented by similar hash values. As a metric for determining the distance between hash values, the Hamming distance is used, this counts the number of dis-tinct bits.To organize the base of hash values, a vp-tree is used, which is an implementation of the binary space partitioning struc-ture.For the experimental study of the methodology, the Caltech-256 data set was used, which contains 30607 images divided into 256 categories, the Difference Hash, P-Hash and Wavelet Hash algorithms were used as perceptual hashing algorithms, the study was carried out in the Google Colab environment.As part of an experimental study, the robustnessof hashing algorithms to modification, compression, blurring, noise, and image rotation was examined. In addition, a study was made of the process of building a vp-tree and the process of searching for images in the tree. As a result of experiments, it was found that each of the hashing algorithms has its own advantages and disadvantages. So, the hashing algorithm based on the difference inadjacentpixel values in the image turned out to be the fastest, but it turned out to be not very robustto modification and image rotation. The P-Hash algorithm, based on the use of the discrete cosine transform, showed better resistance to image blurring, but turned out to be sensitive to image compression. The W-Hash algorithm based on the Haar wavelet transform made it possible to construct the most efficient tree structure and provedto be resistant to image modification and compression.The proposed technique is not recommended for use in general-purpose image retrieval systems; however, it can be useful in searching for images in specialized databases. As ways to improve the methodology, one can note the improvement of the vp-tree structure, as well as the search for a more efficient method of image representation, in addition to perceptual hashing.

PMAL: A Proxy Model Active Learning Approach for Vision Based Industrial Applications

Deep Learning models’ performance strongly correlate with availability of annotated data; however, massive data labeling is laborious, expensive, and error-prone when performed by human experts. Active Learning (AL) effectively handles this challenge by selecting the uncertain samples from unlabeled data collection, but the existing AL approaches involve repetitive human feedback for labeling uncertain samples, thus rendering these techniques infeasible to be deployed in industry related real-world applications. In the proposed Proxy Model based Active Learning technique (PMAL) , this issue is addressed by replacing human oracle with a deep learning model, where human expertise is reduced to label only two small subsets of data for training proxy model and initializing the AL loop. In the PMAL technique, firstly, proxy model is trained with a small subset of labeled data, which subsequently acts as an oracle for annotating uncertain samples. Secondly, active model's training, uncertain samples extraction via uncertainty sampling, and annotation through proxy model is carried out until predefined iterations to achieve higher accuracy and labeled data. Finally, the active model is evaluated using testing data to verify the effectiveness of our technique for practical applications. The correct annotations by the proxy model are ensured by employing the potentials of explainable artificial intelligence. Similarly, emerging vision transformer is used as an active model to achieve maximum accuracy. Experimental results reveal that the proposed method outperforms the state-of-the-art in terms of minimum labeled data usage and improves the accuracy with 2.2%, 2.6%, and 1.35% on Caltech-101, Caltech-256, and CIFAR-10 datasets, respectively. Since the proposed technique offers a highly reasonable solution to exploit huge multimedia data, it can be widely used in different evolutionary industrial domains.