Masked image modeling (MIM) performs strongly in pre-training large vision Transformers (ViTs). However, small models that are critical for real-world applications cannot or only marginally benefit from this pre-training approach. In this paper, we explore distillation techniques to transfer the success of large MIM-based pre-trained models to smaller ones. We systematically study different options in the distillation framework, including distilling targets, losses, input, network regularization, sequential distillation, etc, revealing that: 1) Distilling token relations is more effective than CLS token- and feature-based distillation; 2) An intermediate layer of the teacher network as target perform better than that using the last layer when the depth of the student mismatches that of the teacher; 3) Weak regularization is preferred; etc. With these findings, we achieve significant fine-tuning accuracy improvements over the scratch MIM pre-training on ImageNet-1K classification, using all the ViT-Tiny, ViT-Small, and ViT-base models, with +4.2%/+2.4%/+1.4% gains, respectively. Our TinyMIM model of base size achieves 52.2 mIoU in AE20K semantic segmentation, which is +4.1 higher than the MAE baseline. Our TinyMIM model of tiny size achieves 79.6% top-1 accuracy on ImageNet-1K image classification, which sets a new record for small vision models of the same size and computation budget. This strong performance suggests an alternative way for developing small vision Transformer models, that is, by exploring better training methods rather than introducing inductive biases into architectures as in most previous works. Code is available at https://github.com/OliverRensu/TinyMIM.
Deep learning-based applications have seen a lot of success in recent years. Text, audio, image, and video have all been explored with great success using deep learning approaches. The use of convolutional neural networks (CNN) in computer vision, in particular, has yielded reliable results. In order to achieve these results, a large amount of data is required. However, the dataset cannot always be accessible. Moreover, annotating data can be difficult and time-consuming. Self-training is a semi-supervised approach that managed to alleviate this problem and achieve state-of-the-art performances. Theoretical analysis even proved that it may result in a better generalization than a normal classifier. Another problem neural networks can face is the increasing complexity of modern problems, requiring a high computational and storage cost. One way to mitigate this issue, a strategy that has been inspired by human cognition known as modular learning, can be employed. The principle of the approach is to decompose a complex problem into simpler sub-tasks. This approach has several advantages, including faster learning, better generalization, and enables interpretability. In the first part of this paper, we introduce and evaluate different architectures of modular learning for Dorsal Capsulo-Scapholunate Septum (DCSS) instability classification. Our experiments have shown that modular learning improves performances compared to non-modular systems. Moreover, we found that weighted modular, that is to weight the output using the probabilities from the gating module, achieved an almost perfect classification. In the second part, we present our approach for data labeling and segmentation with self-training applied on shoulder arthroscopy images.
With the flourishing of social media platforms, vision-language pre-training (VLP) recently has received great attention and many remarkable progresses have been achieved. The success of VLP largely benefits from the information complementation and enhancement between different modalities. However, most of recent studies focus on cross-modal contrastive learning (CMCL) to promote image-text alignment by pulling embeddings of positive sample pairs together while pushing those of negative pairs apart, which ignores the natural asymmetry property between different modalities and requires large-scale image-text corpus to achieve arduous progress. To mitigate this predicament, we propose CMAL, a Cross-Modal Associative Learning framework with anchor points detection and cross-modal associative learning for VLP. Specifically, we first respectively embed visual objects and textual tokens into separate hypersphere spaces to learn intra-modal hidden features, and then design a cross-modal associative prompt layer to perform anchor point masking and swap feature filling for constructing a hybrid cross-modal associative prompt. Afterwards, we exploit a unified semantic encoder to learn their cross-modal interactive features for context adaptation. Finally, we design an associative mapping classification layer to learn potential associative mappings between modalities at anchor points, within which we develop a fresh self-supervised associative mapping classification task to boost CMAL's performance. Experimental results verify the effectiveness of CMAL, showing that it achieves competitive performance against previous CMCL-based methods on four common downstream vision-and-language tasks, with significantly fewer corpus. Especially, CMAL obtains new state-of-the-art results on SNLI-VE and REC (testA).
The detection of anomalous tissue regions (ATRs) within affected tissues is crucial in clinical diagnosis and pathological studies. Conventional automated ATR detection methods, primarily based on histology images alone, falter in cases where ATRs and normal tissues have subtle visual differences. The recent spatial transcriptomics (ST) technology profiles gene expressions across tissue regions, offering a molecular perspective for detecting ATRs. However, there is a dearth of ATR detection methods that effectively harness complementary information from both histology images and ST. To address this gap, we propose MEATRD, a novel ATR detection method that integrates histology image and ST data. MEATRD is trained to reconstruct image patches and gene expression profiles of normal tissue spots (inliers) from their multimodal embeddings, followed by learning a one-class classification AD model based on latent multimodal reconstruction errors. This strategy harmonizes the strengths of reconstruction-based and one-class classification approaches. At the heart of MEATRD is an innovative masked graph dual-attention transformer (MGDAT) network, which not only facilitates cross-modality and cross-node information sharing but also addresses the model over-generalization issue commonly seen in reconstruction-based AD methods. Additionally, we demonstrate that modality-specific, task-relevant information is collated and condensed in multimodal bottleneck encoding generated in MGDAT, marking the first theoretical analysis of the informational properties of multimodal bottleneck encoding. Extensive evaluations across eight real ST datasets reveal MEATRD's superior performance in ATR detection, surpassing various state-of-the-art AD methods. Remarkably, MEATRD also proves adept at discerning ATRs that only show slight visual deviations from normal tissues.
Designing discriminative powerful texture features robust to realistic imaging conditions is a challenging computer vision problem with many applications, including material recognition and analysis of satellite or aerial imagery. In the past, most texture description approaches were based on dense orderless statistical distribution of local features. However, most recent approaches to texture recognition and remote sensing scene classification are based on Convolutional Neural Networks (CNNs). The d facto practice when learning these CNN models is to use RGB patches as input with training performed on large amounts of labeled data (ImageNet). In this paper, we show that Binary Patterns encoded CNN models, codenamed TEX-Nets, trained using mapped coded images with explicit texture information provide complementary information to the standard RGB deep models. Additionally, two deep architectures, namely early and late fusion, are investigated to combine the texture and color information. To the best of our knowledge, we are the first to investigate Binary Patterns encoded CNNs and different deep network fusion architectures for texture recognition and remote sensing scene classification. We perform comprehensive experiments on four texture recognition datasets and four remote sensing scene classification benchmarks: UC-Merced with 21 scene categories, WHU-RS19 with 19 scene classes, RSSCN7 with 7 categories and the recently introduced large scale aerial image dataset (AID) with 30 aerial scene types. We demonstrate that TEX-Nets provide complementary information to standard RGB deep model of the same network architecture. Our late fusion TEX-Net architecture always improves the overall performance compared to the standard RGB network on both recognition problems. Our final combination outperforms the state-of-the-art without employing fine-tuning or ensemble of RGB network architectures.
The novel coronavirus universally known as the COVID-19 outbreak arises at the end of 2019 in one of the East Asian countries and it is subjected to widespread discussion and debate. There are almost 200 countries affected across the globe by COVID-19 and it has ruined many lives and the global economy. The virus is spreading very rapidly at the pace of around 10 fold in less than a month. Also, in the case of COVID- 19 it is critical to detect the infection as it employs various symptoms which may differ from person to person. Hence, diagnosis in starting stage and treatment are very much important for such type of infectious disease. The chest x-ray is one of the primary techniques among blood tests and Computed Tomography contributes a major role in the early diagnosis of COVID-19. There is a rising need for automated and auxiliary diagnostic tools for early diagnosis, as there are no accurate and truthful automated tool kits on hand. In this research study, we have designed a Convolution Neural Network architecture a deep net for the classification of x-ray images of chest among two classes: COVID-19 or Non-COVID- 19 infection. The anticipated model is expected to provide accurate diagnostic results and produced classification accuracy of 99%, 100%, and 100% with 70%-30%,75%-25% and 80%-20% train-test data split respectively, for the binary classification of the x-ray image to be COVID-19 or Non-COVID-19 infection category. We have designed the CNN with optimized parameters with 3 convolution layers and optimized number of filters in each layer.
Ensemble methods are arguably the most trustworthy techniques for boosting the performance of machine learning models. Popular independent ensembles (IE) relying on naive averaging/voting scheme have been of typical choice for most applications involving deep neural networks, but they do not consider advanced collaboration among ensemble models. In this paper, we propose new ensemble methods specialized for deep neural networks, called confident multiple choice learning (CMCL): it is a variant of multiple choice learning (MCL) via addressing its overconfidence issue.In particular, the proposed major components of CMCL beyond the original MCL scheme are (i) new loss, i.e., confident oracle loss, (ii) new architecture, i.e., feature sharing and (iii) new training method, i.e., stochastic labeling. We demonstrate the effect of CMCL via experiments on the image classification on CIFAR and SVHN, and the foreground-background segmentation on the iCoseg. In particular, CMCL using 5 residual networks provides 14.05% and 6.60% relative reductions in the top-1 error rates from the corresponding IE scheme for the classification task on CIFAR and SVHN, respectively.
We present a method for metric optimization in the Large Deformation Diffeomorphic Metric Mapping (LDDMM) framework, by treating the induced Riemannian metric on the space of diffeomorphisms as a kernel in a machine learning context. For simplicity, we choose the kernel Fischer Linear Discriminant Analysis (KLDA) as the framework. Optimizing the kernel parameters in an Expectation-Maximization framework, we define model fidelity via the hinge loss of the decision function. The resulting algorithm optimizes the parameters of the LDDMM norm-inducing differential operator as a solution to a group-wise registration and classification problem. In practice, this may lead to a biology-aware registration, focusing its attention on the predictive task at hand such as identifying the effects of disease. We first tested our algorithm on a synthetic dataset, showing that our parameter selection improves registration quality and classification accuracy. We then tested the algorithm on 3D subcortical shapes from the Schizophrenia cohort Schizconnect. Our Schizpohrenia-Control predictive model showed significant improvement in ROC AUC compared to baseline parameters.
Alzheimer's Disease (AD) is the most common type of dementia. In all leading countries, it is one of the primary reasons of death in senior citizens. Currently, it is diagnosed by calculating the MSME score and by the manual study of MRI Scan. Also, different machine learning methods are utilized for automatic diagnosis but existing has some limitations in terms of accuracy. In this paper, we have proposed some novel preprocessing techniques that have significantly increased the accuracy and at the same time decreased the training time of various classification algorithms. First, we have converted the ADNI dataset which was in 4D format into 2D form. We have also mitigated the computation costs by reducing the parameters of the input dataset while preserving important and relevant data. We have achieved this by using different preprocessing steps like grayscale image conversion, Histogram equalization and selective clipping of dataset. We observed a highest accuracy of 97.52% and a sensitivity of 97.6% in our testing dataset.
To reduce the overwhelming size of Deep Neural Networks (DNN) teacher-student methodology tries to transfer knowledge from a complex teacher network to a simple student network. We instead propose a novel method called the teacher-class network consisting of a single teacher and multiple student networks (i.e. class of students). Instead of transferring knowledge to one student only, the proposed method transfers a chunk of knowledge to each student. Our students are not trained for problem-specific logits, they are trained to mimic knowledge (dense representation) learned by the teacher network thus the combined knowledge learned by the class of students can be used to solve other problems as well. The proposed teacher-class architecture is evaluated on several benchmark datasets such as MNIST, Fashion MNIST, IMDB Movie Reviews, CAMVid, CIFAR-10 and ImageNet on multiple tasks including image classification, sentiment classification and segmentation. Our approach outperforms the state of-the-art single student approach in terms of accuracy as well as computational cost while achieving 10-30 times reduction in parameters.
In this paper we introduce a novel, unified, open-source model interpretability library for PyTorch [12]. The library contains generic implementations of a number of gradient and perturbation-based attribution algorithms, also known as feature, neuron and layer importance algorithms, as well as a set of evaluation metrics for these algorithms. It can be used for both classification and non-classification models including graph-structured models built on Neural Networks (NN). In this paper we give a high-level overview of supported attribution algorithms and show how to perform memory-efficient and scalable computations. We emphasize that the three main characteristics of the library are multimodality, extensibility and ease of use. Multimodality supports different modality of inputs such as image, text, audio or video. Extensibility allows adding new algorithms and features. The library is also designed for easy understanding and use. Besides, we also introduce an interactive visualization tool called Captum Insights that is built on top of Captum library and allows sample-based model debugging and visualization using feature importance metrics.
Active learning aims to develop label-efficient algorithms by querying the most representative samples to be labeled by a human annotator. Current active learning techniques either rely on model uncertainty to select the most uncertain samples or use clustering or reconstruction to choose the most diverse set of unlabeled examples. While uncertainty-based strategies are susceptible to outliers, solely relying on sample diversity does not capture the information available on the main task. In this work, we develop a semi-supervised minimax entropy-based active learning algorithm that leverages both uncertainty and diversity in an adversarial manner. Our model consists of an entropy minimizing feature encoding network followed by an entropy maximizing classification layer. This minimax formulation reduces the distribution gap between the labeled/unlabeled data, while a discriminator is simultaneously trained to distinguish the labeled/unlabeled data. The highest entropy samples from the classifier that the discriminator predicts as unlabeled are selected for labeling. We evaluate our method on various image classification and semantic segmentation benchmark datasets and show superior performance over the state-of-the-art methods.
Advancements in deep learning are revolutionizing the classification of remote-sensing images. Transformer-based architectures, utilizing self-attention mechanisms, have emerged as alternatives to conventional convolution methods, enabling the capture of long-range dependencies along with global relationships in the image. Motivated by these advancements, this paper presents StrideNET, a novel dual-branch architecture designed for terrain recognition and implicit properties estimation. The terrain recognition branch utilizes the Swin Transformer, leveraging its hierarchical representation and low computational cost to efficiently capture both local and global features. The terrain properties branch focuses on the extraction of surface properties such as roughness and slipperiness using a statistical texture analysis method. By computing surface terrain properties, an enhanced environmental perception can be obtained. The StrideNET model is trained on a dataset comprising four target terrain classes: Grassy, Marshy, Sandy, and Rocky. StrideNET attains competitive performance compared to contemporary methods. The implications of this work extend to various applications, including environmental monitoring, land use and land cover (LULC) classification, disaster response, precision agriculture, and much more.
Class imbalance is a critical issue in image classification that significantly affects the performance of deep recognition models. In this work, we first identify a network degeneration dilemma that hinders the model learning by introducing a high linear dependence among the features inputted into the classifier. To overcome this challenge, we propose a novel framework called Whitening-Net to mitigate the degenerate solutions, in which ZCA whitening is integrated before the linear classifier to normalize and decorrelate the batch samples. However, in scenarios with extreme class imbalance, the batch covariance statistic exhibits significant fluctuations, impeding the convergence of the whitening operation. Therefore, we propose two covariance-corrected modules, the Group-based Relatively Balanced Batch Sampler (GRBS) and the Batch Embedded Training (BET), to get more accurate and stable batch covariance, thereby reinforcing the capability of whitening. Our modules can be trained end-to-end without incurring substantial computational costs. Comprehensive empirical evaluations conducted on benchmark datasets, including CIFAR-LT-10/100, ImageNet-LT, and iNaturalist-LT, validate the effectiveness of our proposed approaches.
Weight averaging has become a standard technique for enhancing model performance. However, methods such as Stochastic Weight Averaging (SWA) and Latest Weight Averaging (LAWA) often require manually designed procedures to sample from the training trajectory, and the results depend heavily on hyperparameter tuning. To minimize human effort, this paper proposes a simple yet efficient algorithm called Selective Weight Averaging (SeWA), which adaptively selects checkpoints during the final stages of training for averaging. Based on SeWA, we show that only a few points are needed to achieve better generalization and faster convergence. Theoretically, solving the discrete subset selection problem is inherently challenging. To address this, we transform it into a continuous probabilistic optimization framework and employ the Gumbel-Softmax estimator to learn the non-differentiable mask for each checkpoint. Further, we theoretically derive the SeWA's stability-based generalization bounds, which are sharper than that of SGD under both convex and non-convex assumptions. Finally, solid extended experiments in various domains, including behavior cloning, image classification, and text classification, further validate the effectiveness of our approach.
Traditional diagnostic methods like colonoscopy are invasive yet critical tools necessary for accurately diagnosing colorectal cancer (CRC). Detection of CRC at early stages is crucial for increasing patient survival rates. However, colonoscopy is dependent on obtaining adequate and high-quality endoscopic images. Prolonged invasive procedures are inherently risky for patients, while suboptimal or insufficient images hamper diagnostic accuracy. These images, typically derived from video frames, often exhibit similar patterns, posing challenges in discrimination. To overcome these challenges, we propose a novel Deep Learning network built on a Few-Shot Learning architecture, which includes a tailored feature extractor, task interpolation, relational embedding, and a bi-level routing attention mechanism. The Few-Shot Learning paradigm enables our model to rapidly adapt to unseen fine-grained endoscopic image patterns, and the task interpolation augments the insufficient images artificially from varied instrument viewpoints. Our relational embedding approach discerns critical intra-image features and captures inter-image transitions between consecutive endoscopic frames, overcoming the limitations of Convolutional Neural Networks (CNNs). The integration of a light-weight attention mechanism ensures a concentrated analysis of pertinent image regions. By training on diverse datasets, the model's generalizability and robustness are notably improved for handling endoscopic images. Evaluated on Kvasir dataset, our model demonstrated superior performance, achieving an accuracy of 90.1\%, precision of 0.845, recall of 0.942, and an F1 score of 0.891. This surpasses current state-of-the-art methods, presenting a promising solution to the challenges of invasive colonoscopy by optimizing CRC detection through advanced image analysis.
Image data has a great potential of helping conventional visual inspections of civil engineering structures due to the ease of data acquisition and the advantages in capturing visual information. A variety of techniques have been proposed to detect damages, such as cracks and spalling on a close-up image of a single component (columns and road surfaces etc.). However, these techniques commonly suffer from severe false-positives especially when the image includes multiple components of different structures. To reduce the false-positives and extract reliable information about the structures' conditions, detection and localization of critical structural components are important first steps preceding the damage assessment. This study aims at recognizing bridge structural and non-structural components from images of urban scenes. During the bridge component recognition, every image pixel is classified into one of the five classes (non-bridge, columns, beams and slabs, other structural, other nonstructural) by multi-scale convolutional neural networks (multi-scale CNNs). To reduce false-positives and get consistent labels, the component classifications are integrated with scene understanding by an additional classifier with 10 higher-level scene classes (building, greenery, person, pavement, signs and poles, vehicles, bridges, water, sky, and others). The bridge component recognition integrated with the scene understanding is compared with the naive approach without scene classification in terms of accuracy, false-positives and consistencies to demonstrate the effectiveness of the integrated approach.
Visible-Infrared person Re-IDentification (VI-ReID) is a challenging cross-modality image retrieval task that aims to match pedestrians' images across visible and infrared cameras. To solve the modality gap, existing mainstream methods adopt a learning paradigm converting the image retrieval task into an image classification task with cross-entropy loss and auxiliary metric learning losses. These losses follow the strategy of adjusting the distribution of extracted embeddings to reduce the intra-class distance and increase the inter-class distance. However, such objectives do not precisely correspond to the final test setting of the retrieval task, resulting in a new gap at the optimization level. By rethinking these keys of VI-ReID, we propose a simple and effective method, the Multi-level Cross-modality Joint Alignment (MCJA), bridging both modality and objective-level gap. For the former, we design the Modality Alignment Augmentation, which consists of three novel strategies, the weighted grayscale, cross-channel cutmix, and spectrum jitter augmentation, effectively reducing modality discrepancy in the image space. For the latter, we introduce a new Cross-Modality Retrieval loss. It is the first work to constrain from the perspective of the ranking list, aligning with the goal of the testing stage. Moreover, based on the global feature only, our method exhibits good performance and can serve as a strong baseline method for the VI-ReID community.
Person re-identification aims at the maintenance of a global identity as a person moves among non-overlapping surveillance cameras. It is a hard task due to different illumination conditions, viewpoints and the small number of annotated individuals from each pair of cameras (small-sample-size problem). Collaborative Representation based Classification (CRC) has been employed successfully to address the small-sample-size problem in computer vision. However, the original CRC formulation is not well-suited for person re-identification since it does not consider that probe and gallery samples are from different cameras. Furthermore, it is a linear model, while appearance changes caused by different camera conditions indicate a strong nonlinear transition between cameras. To overcome such limitations, we propose the Kernel Cross-View Collaborative Representation based Classification (Kernel X-CRC) that represents probe and gallery images by balancing representativeness and similarity nonlinearly. It assumes that a probe and its corresponding gallery image are represented with similar coding vectors using individuals from the training set. Experimental results demonstrate that our assumption is true when using a high-dimensional feature vector and becomes more compelling when dealing with a low-dimensional and discriminative representation computed using a common subspace learning method. We achieve state-of-the-art for rank-1 matching rates in two person re-identification datasets (PRID450S and GRID) and the second best results on VIPeR and CUHK01 datasets.
Transfer learning using pre-trained Convolutional Neural Networks (CNNs) has been successfully applied to images for different classification tasks. In this paper, we propose a new pipeline for pain expression recognition in neonates using transfer learning. Specifically, we propose to exploit a pre-trained CNN that was originally trained on a relatively similar dataset for face recognition (VGG Face) as well as CNNs that were pre-trained on a relatively different dataset for image classification (iVGG F,M, and S) to extract deep features from neonates' faces. In the final stage, several supervised machine learning classifiers are trained to classify neonates' facial expression into pain or no pain expression. The proposed pipeline achieved, on a testing dataset, 0.841 AUC and 90.34 accuracy, which is approx. 7 higher than the accuracy of handcrafted traditional features. We also propose to combine deep features with traditional features and hypothesize that the mixed features would improve pain classification performance. Combining deep features with traditional features achieved 92.71 accuracy and 0.948 AUC. These results show that transfer learning, which is a faster and more practical option than training CNN from the scratch, can be used to extract useful features for pain expression recognition in neonates. It also shows that combining deep features with traditional handcrafted features is a good practice to improve the performance of pain expression recognition and possibly the performance of similar applications.