Part-based representations have been shown to be very useful for image classification. Learning part-based models is often viewed as a two-stage problem. First, a collection of informative parts is discovered, using heuristics that promote part distinctiveness and diversity, and then classifiers are trained on the vector of part responses. In this paper we unify the two stages and learn the image classifiers and a set of shared parts jointly. We generate an initial pool of parts by randomly sampling part candidates and selecting a good subset using L1/L2 regularization. All steps are driven "directly" by the same objective namely the classification loss on a training set. This lets us do away with engineered heuristics. We also introduce the notion of "negative parts", intended as parts that are negatively correlated with one or more classes. Negative parts are complementary to the parts discovered by other methods, which look only for positive correlations.
The present study is an attempt made to create a new tool for the analysis of Land Use Land Cover (LUCL) images in 3D visualization. This study mainly uses spatial data mining techniques on high resolution LULC satellite imagery. Visualization of feature space allows exploration of patterns in the image data and insight into the classification process and related uncertainty. Visual Data Mining provides added value to image classifications as the user can be involved in the classification process providing increased confidence in and understanding of the results. In this study, we present a prototype of image segmentation, K-Means clustering and 3D visualization tool for visual data mining (VDM) of LUCL satellite imagery into volume visualization. This volume based representation divides feature space into spheres or voxels. The visualization tool is showcased in a classification study of high-resolution LULC imagery of Latur district (Maharashtra state, India) is used as sample data.
This paper explores a hierarchical prompting mechanism for the hierarchical image classification (HIC) task. Different from prior HIC methods, our hierarchical prompting is the first to explicitly inject ancestor-class information as a tokenized hint that benefits the descendant-class discrimination. We think it well imitates human visual recognition, i.e., humans may use the ancestor class as a prompt to draw focus on the subtle differences among descendant classes. We model this prompting mechanism into a Transformer with Hierarchical Prompting (TransHP). TransHP consists of three steps: 1) learning a set of prompt tokens to represent the coarse (ancestor) classes, 2) on-the-fly predicting the coarse class of the input image at an intermediate block, and 3) injecting the prompt token of the predicted coarse class into the intermediate feature. Though the parameters of TransHP maintain the same for all input images, the injected coarse-class prompt conditions (modifies) the subsequent feature extraction and encourages a dynamic focus on relatively subtle differences among the descendant classes. Extensive experiments show that TransHP improves image classification on accuracy (e.g., improving ViT-B/16 by +2.83% ImageNet classification accuracy), training data efficiency (e.g., +12.69% improvement under 10% ImageNet training data), and model explainability. Moreover, TransHP also performs favorably against prior HIC methods, showing that TransHP well exploits the hierarchical information. The code is available at: https://github.com/WangWenhao0716/TransHP.
Artificial Intelligence (XAI) has found numerous applications in computer vision. While image classification-based explainability techniques have garnered significant attention, their counterparts in semantic segmentation have been relatively neglected. Given the prevalent use of image segmentation, ranging from medical to industrial deployments, these techniques warrant a systematic look. In this paper, we present the first comprehensive survey on XAI in semantic image segmentation. This work focuses on techniques that were either specifically introduced for dense prediction tasks or were extended for them by modifying existing methods in classification. We analyze and categorize the literature based on application categories and domains, as well as the evaluation metrics and datasets used. We also propose a taxonomy for interpretable semantic segmentation, and discuss potential challenges and future research directions.
Convolutional neural networks (CNNs) achieve prevailing results in segmentation tasks nowadays and represent the state-of-the-art for image-based analysis. However, the understanding of the accurate decision-making process of a CNN is rather unknown. The research area of explainable artificial intelligence (xAI) primarily revolves around understanding and interpreting this black-box behavior. One way of interpreting a CNN is the use of class activation maps (CAMs) that represent heatmaps to indicate the importance of image areas for the prediction of the CNN. For classification tasks, a variety of CAM algorithms exist. But for segmentation tasks, only one CAM algorithm for the interpretation of the output of a CNN exist. We propose a transfer between existing classification- and segmentation-based methods for more detailed, explainable, and consistent results which show salient pixels in semantic segmentation tasks. The resulting Seg-HiRes-Grad CAM is an extension of the segmentation-based Seg-Grad CAM with the transfer to the classification-based HiRes CAM. Our method improves the previously-mentioned existing segmentation-based method by adjusting it to recently published classification-based methods. Especially for medical image segmentation, this transfer solves existing explainability disadvantages.
The majority of primary Central Nervous System (CNS) tumors in the brain are among the most aggressive diseases affecting humans. Early detection of brain tumor types, whether benign or malignant, glial or non-glial, is critical for cancer prevention and treatment, ultimately improving human life expectancy. Magnetic Resonance Imaging (MRI) stands as the most effective technique to detect brain tumors by generating comprehensive brain images through scans. However, human examination can be error-prone and inefficient due to the complexity, size, and location variability of brain tumors. Recently, automated classification techniques using machine learning (ML) methods, such as Convolutional Neural Network (CNN), have demonstrated significantly higher accuracy than manual screening, while maintaining low computational costs. Nonetheless, deep learning-based image classification methods, including CNN, face challenges in estimating class probabilities without proper model calibration. In this paper, we propose a novel brain tumor image classification method, called SIBOW-SVM, which integrates the Bag-of-Features (BoF) model with SIFT feature extraction and weighted Support Vector Machines (wSVMs). This new approach effectively captures hidden image features, enabling the differentiation of various tumor types and accurate label predictions. Additionally, the SIBOW-SVM is able to estimate the probabilities of images belonging to each class, thereby providing high-confidence classification decisions. We have also developed scalable and parallelable algorithms to facilitate the practical implementation of SIBOW-SVM for massive images. As a benchmark, we apply the SIBOW-SVM to a public data set of brain tumor MRI images containing four classes: glioma, meningioma, pituitary, and normal. Our results show that the new method outperforms state-of-the-art methods, including CNN.
In recent years, Vision Transformers (ViTs) have shown promising classification performance over Convolutional Neural Networks (CNNs) due to their self-attention mechanism. Many researchers have incorporated ViTs for Hyperspectral Image (HSI) classification. HSIs are characterised by narrow contiguous spectral bands, providing rich spectral data. Although ViTs excel with sequential data, they cannot extract spectral-spatial information like CNNs. Furthermore, to have high classification performance, there should be a strong interaction between the HSI token and the class (CLS) token. To solve these issues, we propose a 3D-Convolution guided Spectral-Spatial Transformer (3D-ConvSST) for HSI classification that utilizes a 3D-Convolution Guided Residual Module (CGRM) in-between encoders to "fuse" the local spatial and spectral information and to enhance the feature propagation. Furthermore, we forego the class token and instead apply Global Average Pooling, which effectively encodes more discriminative and pertinent high-level features for classification. Extensive experiments have been conducted on three public HSI datasets to show the superiority of the proposed model over state-of-the-art traditional, convolutional, and Transformer models. The code is available at https://github.com/ShyamVarahagiri/3D-ConvSST.
Masked Autoencoder (MAE) has recently been shown to be effective in pre-training Vision Transformers (ViT) for natural image analysis. By reconstructing full images from partially masked inputs, a ViT encoder aggregates contextual information to infer masked image regions. We believe that this context aggregation ability is particularly essential to the medical image domain where each anatomical structure is functionally and mechanically connected to other structures and regions. Because there is no ImageNet-scale medical image dataset for pre-training, we investigate a self pre-training paradigm with MAE for medical image analysis tasks. Our method pre-trains a ViT on the training set of the target data instead of another dataset. Thus, self pre-training can benefit more scenarios where pre-training data is hard to acquire. Our experimental results show that MAE self pre-training markedly improves diverse medical image tasks including chest X-ray disease classification, abdominal CT multi-organ segmentation, and MRI brain tumor segmentation. Code is available at https://github.com/cvlab-stonybrook/SelfMedMAE
AUCMEDI is a Python-based framework for medical image classification. In this paper, we evaluate the capabilities of AUCMEDI, by applying it to multiple datasets. Datasets were specifically chosen to cover a variety of medical disciplines and imaging modalities. We designed a simple pipeline using Jupyter notebooks and applied it to all datasets. Results show that AUCMEDI was able to train a model with accurate classification capabilities for each dataset: Averaged AUC per dataset range between 0.82 and 1.0, averaged F1 scores range between 0.61 and 1.0. With its high adaptability and strong performance, AUCMEDI proves to be a powerful instrument to build widely applicable neural networks. The notebooks serve as application examples for AUCMEDI.
Over the past few years, there has been significant interest in Quantum Machine Learning (QML) among researchers, as it has the potential to transform the field of machine learning. Several models that exploit the properties of quantum mechanics have been developed for practical applications. In this study, we investigated the application of our previously proposed quantum pre-processing filter (QPF) to binary image classification. We evaluated the QPF on four datasets: MNIST (handwritten digits), EMNIST (handwritten digits and alphabets), CIFAR-10 (photographic images) and GTSRB (real-life traffic sign images). Similar to our previous multi-class classification results, the application of QPF improved the binary image classification accuracy using neural network against MNIST, EMNIST, and CIFAR-10 from 98.9% to 99.2%, 97.8% to 98.3%, and 71.2% to 76.1%, respectively, but degraded it against GTSRB from 93.5% to 92.0%. We then applied QPF in cases using a smaller number of training and testing samples, i.e. 80 and 20 samples per class, respectively. In order to derive statistically stable results, we conducted the experiment with 100 trials choosing randomly different training and testing samples and averaging the results. The result showed that the application of QPF did not improve the image classification accuracy against MNIST and EMNIST but improved it against CIFAR-10 and GTSRB from 65.8% to 67.2% and 90.5% to 91.8%, respectively. Further research will be conducted as part of future work to investigate the potential of QPF to assess the scalability of the proposed approach to larger and complex datasets.
Convolutional Neural Networks (CNN) has achieved a great success in image recognition task by automatically learning a hierarchical feature representation from raw data. While the majority of Time-Series Classification (TSC) literature is focused on 1D signals, this paper uses Recurrence Plots (RP) to transform time-series into 2D texture images and then take advantage of the deep CNN classifier. Image representation of time-series introduces different feature types that are not available for 1D signals, and therefore TSC can be treated as texture image recognition task. CNN model also allows learning different levels of representations together with a classifier, jointly and automatically. Therefore, using RP and CNN in a unified framework is expected to boost the recognition rate of TSC. Experimental results on the UCR time-series classification archive demonstrate competitive accuracy of the proposed approach, compared not only to the existing deep architectures, but also to the state-of-the art TSC algorithms.
Event detection is an important step in extracting knowledge from the video. In this paper, we propose a deep learning approach to detect events in a soccer match emphasizing the distinction between images of red and yellow cards and the correct detection of the images of selected events from other images. This method includes the following three modules: i) the variational autoencoder (VAE) module to differentiate between soccer images and others image, ii) the image classification module to classify the images of events, and iii) the fine-grain image classification module to classify the images of red and yellow cards. Additionally, a new dataset was introduced for soccer images classification that is employed to train the networks mentioned in the paper. In the final section, 10 UEFA Champions League matches are used to evaluate the networks' performance and precision in detecting the events. The experiments demonstrate that the proposed method achieves better performance than state-of-the-art methods.
Deep learning models for image classification are often trained at a resolution of 224 x 224 pixels for historical and efficiency reasons. However, chest X-rays are acquired at a much higher resolution to display subtle pathologies. This study investigates the effect of training resolution on chest X-ray classification performance, using the chest X-ray 14 dataset. The results show that training with a higher image resolution, specifically 1024 x 1024 pixels, results in the best overall classification performance with a mean AUC of 84.2 % compared to 82.7 % when trained with 256 x 256 pixel images. Additionally, comparison of bounding boxes and GradCAM saliency maps suggest that low resolutions, such as 256 x 256 pixels, are insufficient for identifying small pathologies and force the model to use spurious discriminating features. Our code is publicly available at https://gitlab.lrz.de/IP/cxr-resolution
Despite the progress made in the field of medical imaging, it remains a large area of open research, especially due to the variety of imaging modalities and disease-specific characteristics. This paper is a comparative study describing the potential of using local binary patterns (LBP), deep features and the bag-of-visual words (BoVW) scheme for the classification of histopathological images. We introduce a new dataset, \emph{KIMIA Path960}, that contains 960 histopathology images belonging to 20 different classes (different tissue types). We make this dataset publicly available. The small size of the dataset and its inter- and intra-class variability makes it ideal for initial investigations when comparing image descriptors for search and classification in complex medical imaging cases like histopathology. We investigate deep features, LBP histograms and BoVW to classify the images via leave-one-out validation. The accuracy of image classification obtained using LBP was 90.62\% while the highest accuracy using deep features reached 94.72\%. The dictionary approach (BoVW) achieved 96.50\%. Deep solutions may be able to deliver higher accuracies but they need extensive training with a large number of (balanced) image datasets.
In an effort to catalog insect biodiversity, we propose a new large dataset of hand-labelled insect images, the BIOSCAN-Insect Dataset. Each record is taxonomically classified by an expert, and also has associated genetic information including raw nucleotide barcode sequences and assigned barcode index numbers, which are genetically-based proxies for species classification. This paper presents a curated million-image dataset, primarily to train computer-vision models capable of providing image-based taxonomic assessment, however, the dataset also presents compelling characteristics, the study of which would be of interest to the broader machine learning community. Driven by the biological nature inherent to the dataset, a characteristic long-tailed class-imbalance distribution is exhibited. Furthermore, taxonomic labelling is a hierarchical classification scheme, presenting a highly fine-grained classification problem at lower levels. Beyond spurring interest in biodiversity research within the machine learning community, progress on creating an image-based taxonomic classifier will also further the ultimate goal of all BIOSCAN research: to lay the foundation for a comprehensive survey of global biodiversity. This paper introduces the dataset and explores the classification task through the implementation and analysis of a baseline classifier.
The Jacobian-based Saliency Map Attack is a family of adversarial attack methods for fooling classification models, such as deep neural networks for image classification tasks. By saturating a few pixels in a given image to their maximum or minimum values, JSMA can cause the model to misclassify the resulting adversarial image as a specified erroneous target class. We propose two variants of JSMA, one which removes the requirement to specify a target class, and another that additionally does not need to specify whether to only increase or decrease pixel intensities. Our experiments highlight the competitive speeds and qualities of these variants when applied to datasets of hand-written digits and natural scenes.
Identifying and reconstructing what we see from brain activity gives us a special insight into investigating how the biological visual system represents the world. While recent efforts have achieved high-performance image classification and high-quality image reconstruction from brain signals collected by Functional Magnetic Resonance Imaging (fMRI) or magnetoencephalogram (MEG), the expensiveness and bulkiness of these devices make relevant applications difficult to generalize to practical applications. On the other hand, Electroencephalography (EEG), despite its advantages of ease of use, cost-efficiency, high temporal resolution, and non-invasive nature, has not been fully explored in relevant studies due to the lack of comprehensive datasets. To address this gap, we introduce EEG-ImageNet, a novel EEG dataset comprising recordings from 16 subjects exposed to 4000 images selected from the ImageNet dataset. EEG-ImageNet consists of 5 times EEG-image pairs larger than existing similar EEG benchmarks. EEG-ImageNet is collected with image stimuli of multi-granularity labels, i.e., 40 images with coarse-grained labels and 40 with fine-grained labels. Based on it, we establish benchmarks for object classification and image reconstruction. Experiments with several commonly used models show that the best models can achieve object classification with accuracy around 60% and image reconstruction with two-way identification around 64%. These results demonstrate the dataset's potential to advance EEG-based visual brain-computer interfaces, understand the visual perception of biological systems, and provide potential applications in improving machine visual models.
Single image-level annotations only correctly describe an often small subset of an image's content, particularly when complex real-world scenes are depicted. While this might be acceptable in many classification scenarios, it poses a significant challenge for applications where the set of classes differs significantly between training and test time. In this paper, we take a closer look at the implications in the context of $\textit{few-shot learning}$. Splitting the input samples into patches and encoding these via the help of Vision Transformers allows us to establish semantic correspondences between local regions across images and independent of their respective class. The most informative patch embeddings for the task at hand are then determined as a function of the support set via online optimization at inference time, additionally providing visual interpretability of `$\textit{what matters most}$' in the image. We build on recent advances in unsupervised training of networks via masked image modelling to overcome the lack of fine-grained labels and learn the more general statistical structure of the data while avoiding negative image-level annotation influence, $\textit{aka}$ supervision collapse. Experimental results show the competitiveness of our approach, achieving new state-of-the-art results on four popular few-shot classification benchmarks for $5$-shot and $1$-shot scenarios.
The identification of lesion within medical image data is necessary for diagnosis, treatment and prognosis. Segmentation and classification approaches are mainly based on supervised learning with well-paired image-level or voxel-level labels. However, labeling the lesion in medical images is laborious requiring highly specialized knowledge. We propose a medical image synthesis model named abnormal-to-normal translation generative adversarial network (ANT-GAN) to generate a normal-looking medical image based on its abnormal-looking counterpart without the need for paired training data. Unlike typical GANs, whose aim is to generate realistic samples with variations, our more restrictive model aims at producing a normal-looking image corresponding to one containing lesions, and thus requires a special design. Being able to provide a "normal" counterpart to a medical image can provide useful side information for medical imaging tasks like lesion segmentation or classification validated by our experiments. In the other aspect, the ANT-GAN model is also capable of producing highly realistic lesion-containing image corresponding to the healthy one, which shows the potential in data augmentation verified in our experiments.
We introduce MultiDepth, a novel training strategy and convolutional neural network (CNN) architecture that allows approaching single-image depth estimation (SIDE) as a multi-task problem. SIDE is an important part of road scene understanding. It, thus, plays a vital role in advanced driver assistance systems and autonomous vehicles. Best results for the SIDE task so far have been achieved using deep CNNs. However, optimization of regression problems, such as estimating depth, is still a challenging task. For the related tasks of image classification and semantic segmentation, numerous CNN-based methods with robust training behavior have been proposed. Hence, in order to overcome the notorious instability and slow convergence of depth value regression during training, MultiDepth makes use of depth interval classification as an auxiliary task. The auxiliary task can be disabled at test-time to predict continuous depth values using the main regression branch more efficiently. We applied MultiDepth to road scenes and present results on the KITTI depth prediction dataset. In experiments, we were able to show that end-to-end multi-task learning with both, regression and classification, is able to considerably improve training and yield more accurate results.