Limited Annotation Research Articles

Quantitative assessment of cracks in concrete structures using active-learning-integrated transformer and unmanned robotic platform

Quantitative assessment of cracks in concrete bridges is crucial for structural health monitoring and digital twin. However, the training of crack segmentation models relies heavily on annotation resources, and their segmentation capabilities are often unsatisfactory in terms of the accuracy of boundary location of thin cracks encountered in practice. In this paper, an active-learning-integrated crack segmentation transformer (ACS-Former) framework is proposed to maximize segmentation performance with limited annotation resources. The two-branch ACS-Former includes (1) a feature pyramid transformer (FPT) for multi-scale crack segmentation and (2) boundary difficulty-aware active learning (BDAL) to select informative images for labeling and incorporation into FPT training. Additionally, an adhesive climbing robot is proposed for image collection of hard-to-access components of large bridges. The on-site operational feasibility and practicability of the proposed ACS-Former and climbing robot are demonstrated by field experiments performed on in-service bridges, including the quantification of cracks narrower than 0.2 mm, as required by engineering codes.

Learning to learn for few-shot continual active learning

Continual learning strives to ensure stability in solving previously seen tasks while demonstrating plasticity in a novel domain. Recent advances in continual learning are mostly confined to a supervised learning setting, especially in NLP domain. In this work, we consider a few-shot continual active learning setting where labeled data are inadequate, and unlabeled data are abundant but with a limited annotation budget. We exploit meta-learning and propose a method, called Meta-Continual Active Learning. This method sequentially queries the most informative examples from a pool of unlabeled data for annotation to enhance task-specific performance and tackles continual learning problems through a meta-objective. Specifically, we employ meta-learning and experience replay to address inter-task confusion and catastrophic forgetting. We further incorporate textual augmentations to avoid memory over-fitting caused by experience replay and sample queries, thereby ensuring generalization. We conduct extensive experiments on benchmark text classification datasets from diverse domains to validate the feasibility and effectiveness of meta-continual active learning. We also analyze the impact of different active learning strategies on various meta continual learning models. The experimental results demonstrate that introducing randomness into sample selection is the best default strategy for maintaining generalization in meta-continual learning framework.

Archaea in the Human Microbiome and Potential Effects on Human Infectious Disease.

Archaea represent a separate domain of life, next to bacteria and eukarya. As components of the human microbiome, archaea have been associated with various diseases, including periodontitis, endodontic infections, small intestinal bacterial overgrowth, and urogenital tract infections. Archaea are generally considered nonpathogenic; the reasons are speculative because of limited knowledge and gene annotation challenges. Nevertheless, archaeal syntrophic principles that shape global microbial networks aid both archaea and potentially pathogenic bacteria. Evaluating archaea interactions remains challenging, requiring clinical studies on inflammatory potential and the effects of archaeal metabolism. Establishing a culture collection is crucial for investigating archaea functions within the human microbiome, which could improve health outcomes in infectious diseases. We summarize potential reasons for archaeal nonpathogenicity, assess the association with infectious diseases in humans, and discuss the necessary experimental steps to enable mechanistic studies involving archaea.

CRISPR-Cas9 knockout screen informs efficient reduction of the Komagataella phaffii secretome

BackgroundThe yeast Komagataella phaffii is widely used for manufacturing recombinant proteins, but secreted titers of recombinant proteins could be improved by genetic engineering. In this study, we hypothesized that cellular resources could be redirected from production of endogenous proteins to production of recombinant proteins by deleting unneeded endogenous proteins. In non-model microorganisms such as K. phaffii, however, genetic engineering is limited by lack gene annotation and knowledge of gene essentiality.ResultsWe identified a set of endogenous secreted proteins in K. phaffii by mass spectrometry and signal peptide prediction. Our efforts to disrupt these genes were hindered by limited annotation of essential genes. To predict essential genes, therefore, we designed, transformed, and sequenced a pooled library of guide RNAs for CRISPR-Cas9-mediated knockout of all endogenous secreted proteins. We then used predicted gene essentiality to guide iterative disruptions of up to 11 non-essential genes. Engineered strains exhibited a ~20× increase in the production of human serum albumin and a twofold increase in the production of a monoclonal antibody.ConclusionsWe demonstrated that disruption of as few as six genes can increase production of recombinant proteins. Further reduction of the endogenous proteome of K. phaffii may further improve strain performance. The pooled library of secretome-targeted guides for CRISPR-Cas9 and knowledge of gene essentiality reported here will facilitate future efforts to engineer K. phaffii for production of other recombinant proteins and enzymes.

MSAug: Multi-Strategy Augmentation for rare classes in semantic segmentation of remote sensing images

Recently, remote sensing images have been widely used in many scenarios, gradually becoming the focus of social attention. Nevertheless, the limited annotation of scarce classes severely reduces segmentation performance. This phenomenon is more prominent in remote sensing image segmentation. Given this, we focus on image fusion and model feedback, proposing a multi-strategy method called MSAug to address the remote sensing imbalance problem. Firstly, we crop rare class images multiple times based on prior knowledge at the image patch level to provide more balanced samples. Secondly, we design an adaptive image enhancement module at the model feedback level to accurately classify rare classes at each stage and dynamically paste and mask different classes to further improve the model’s recognition capabilities. The MSAug method is highly flexible and can be plug-and-play. Experimental results on remote sensing image segmentation datasets show that adding MSAug to any remote sensing image semantic segmentation network can bring varying degrees of performance improvement.

Low-resource entity resolution with domain generalization and active learning

Entity Resolution (ER), a fundamental task in data cleaning and integration, is critical in various fields such as healthcare, e-commerce, and social networks. Traditional ER methods are constrained by the need for substantial labeled samples and the challenge of generalization to unseen domains. To address the low-resource challenge in ER, a novel two-phase framework is proposed. Initially, we introduce the Domain Generalization Entity Resolution (DGER) framework, combining domain adversarial learning and simulated target learning to improve the generalization performance on unseen target domains. Subsequently, to further adapt to the target dataset, we present a novel active learning approach called Domain-Aware Uncertainty Active Learning (DUAL), for fine-tuning the DGER model with minimal annotation cost. DUAL manually annotates target domain samples that are highly uncertain and exhibit high divergence from the source, while assigning pseudo-labels to high-confidence samples. Experimental results on multiple real-world datasets demonstrate that our framework outperforms traditional ER methods in generalizing to unseen domains. Specifically, our DGER method outperforms the best-performing ER baseline in each task, achieving an average F1 score improvement of 9.02% across eight different test tasks. Moreover, within a limited annotation budget during the active learning phase, our DUAL fine-tuning strategy for the ER model outperforms uncertainty-based active learning techniques.

Named Entity Recognition for Chinese Texts on Marine Coral Reef Ecosystems Based on the BERT-BiGRU-Att-CRF Model

In addressing the challenges of non-standardization and limited annotation resources in Chinese marine domain texts, particularly with complex entities like long and nested entities in coral reef ecosystem-related texts, existing Named Entity Recognition (NER) methods often fail to capture deep semantic features, leading to inefficiencies and inaccuracies. This study introduces a deep learning model that integrates Bidirectional Encoder Representations from Transformers (BERT), Bidirectional Gated Recurrent Units (BiGRU), and Conditional Random Fields (CRF), enhanced by an attention mechanism, to improve the recognition of complex entity structures. The model utilizes BERT to capture context-relevant character vectors, employs BiGRU to extract global semantic features, incorporates an attention mechanism to focus on key information, and uses CRF to produce optimized label sequences. We constructed a specialized coral reef ecosystem corpus to evaluate the model’s performance through a series of experiments. The results demonstrated that our model achieved an F1 score of 86.54%, significantly outperforming existing methods. The contributions of this research are threefold: (1) We designed an efficient named entity recognition framework for marine domain texts, improving the recognition of long and nested entities. (2) By introducing the attention mechanism, we enhanced the model’s ability to recognize complex entity structures in coral reef ecosystem texts. (3) This work offers new tools and perspectives for marine domain knowledge graph construction and study, laying a foundation for future research. These advancements propel the development of marine domain text analysis technology and provide valuable references for related research fields.

Self-supervised class-balanced active learning with uncertainty-mastery fusion

Deep active learning (DeepAL) offers a viable solution for enhancing the predictive performance of models with limited annotation costs. Popular DeepAL pipelines fail to (1) adequately and simultaneously draw on unlabeled and labeled samples to learn accurate data representations, and (2) progressively balance the distribution of selected samples across classes during the query. In this paper, we introduce a self-supervised DeepAL (SSAL) framework as a solution to the aforementioned challenges. First, SSAL simultaneously utilizes both labeled data and unlabeled data to learn high-quality representations. The method learns the image rotation degree for unlabeled data in a self-supervised manner, which can be jointly processed with the supervised learning. Second, the sample information is interpreted by fusing the discriminant uncertainty and the presumed mastery ability of each sample. This approach prevents the selection of many nonrepresentative boundary samples based solely on uncertainty and avoids the selection of aggregated samples based solely on clustering. Third, we propose a new sample selection strategy named class rebalanced group sampling. This strategy rebalances classes by the optimal size of query samples for each class, which is induced from the approximation between the population and the model distribution. Experiments were conducted on five public image datasets, namely CIFAR-10/100, FashionMNIST, SVHN, and TinyImageNet, and the results demonstrated the effectiveness of our approach. The code is available at https://github.com/FanSmale/SSAL.

Enhancing Semi-Supervised Few-Shot Hyperspectral Image Classification via Progressive Sample Selection

Hyperspectral images (HSIs) provide valuable spatial–spectral information for ground analysis. However, in few-shot (FS) scenarios, the limited availability of training samples poses significant challenges in capturing the sample distribution under diverse environmental conditions. Semi-supervised learning has shown promise in exploring the distribution of unlabeled samples through pseudo-labels. Nonetheless, FS HSI classification encounters the issue of high intra-class spectral variability and inter-class spectral similarity, which often lead to the diffusion of unreliable pseudo-labels during the iterative process. In this paper, we propose a simple yet effective progressive pseudo-label selection strategy that leverages the spatial–spectral consistency of HSI pixel samples. By leveraging spatially aligned ground materials as connected regions with the same semantic and similar spectrum, pseudo-labeled samples were selected based on round-wise confidence scores. Samples within both spatially and semantically connected regions of FS samples were assigned pseudo-labels and joined subsequent training rounds. Moreover, considering the spatial positions of FS samples that may appear in diverse patterns, to fully utilize unlabeled samples that fall outside the neighborhood of FS samples but still belong to certain connected regions, we designed a matching active learning approach for expert annotation based on the temporal confidence difference. We identified samples with the highest training value in specific regions, utilizing the consistency between predictive labels and expert labels to decide whether to include the region or the sample itself in the subsequent semi-supervised iteration. Experiments on both classic and more recent HSI datasets demonstrated that the proposed base model achieved SOTA performance even with extremely rare labeled samples. Moreover, the extended version with active learning further enhances performance by involving limited additional annotation.

Research of news text classification method based on hierarchical semantics and prior correction

News text is an important branch of natural language processing. Compared to ordinary texts, news text has significant economic and scientific value. The characteristics of news text include structural hierarchy, diverse label categories, and limited high-quality annotation samples. Many machine learning and deep learning methods exist to analyze various forms of news text. However, due to label imbalance, hierarchical semantics, and confusing labels, current methods have limitations. Therefore, this paper proposes a news text classification framework based on hierarchical semantics and prior correction (HSPC). Firstly, data augmentation is used to enhance the diversity of the training set and adversarial learning is employed to improve the resistance of the model with its robustness. Then, a hierarchical feature extraction approach is employed to extract semantic features from different levels of news texts. Consequentially, a feature fusion method is designed to allow the model to focus on relevant hierarchical semantics for label classification. Finally, highly confusing label predictions are corrected to optimize the label prediction of the model and improve confidence. Multiple experiments are performed on four widely used public datasets. The experimental results indicate that HSPC achieves higher classification accuracy compared to other models. On the FCT, AGNews, THUCNews, and Ohsumed datasets, HSPC improves the accuracy by 1.03%, 1.38%, 2.55%, and 1.15%, respectively, compared to state-of-the-art methods. This validates the rationality and effectiveness of the designed mechanisms.

Self-supervised learning for medical image data with anatomy-oriented imaging planes

Self-supervised learning has emerged as a powerful tool for pretraining deep networks on unlabeled data, prior to transfer learning of target tasks with limited annotation. The relevance between the pretraining pretext and target tasks is crucial to the success of transfer learning. Various pretext tasks have been proposed to utilize properties of medical image data (e.g., three dimensionality), which are more relevant to medical image analysis than generic ones for natural images. However, previous work rarely paid attention to data with anatomy-oriented imaging planes, e.g., standard cardiac magnetic resonance imaging views. As these imaging planes are defined according to the anatomy of the imaged organ, pretext tasks effectively exploiting this information can pretrain the networks to gain knowledge on the organ of interest. In this work, we propose two complementary pretext tasks for this group of medical image data based on the spatial relationship of the imaging planes. The first is to learn the relative orientation between the imaging planes and implemented as regressing their intersecting lines. The second exploits parallel imaging planes to regress their relative slice locations within a stack. Both pretext tasks are conceptually straightforward and easy to implement, and can be combined in multitask learning for better representation learning. Thorough experiments on two anatomical structures (heart and knee) and representative target tasks (semantic segmentation and classification) demonstrate that the proposed pretext tasks are effective in pretraining deep networks for remarkably boosted performance on the target tasks, and superior to other recent approaches.

Meibomian glands segmentation in infrared images with limited annotation.

To investigate a pioneering framework for the segmentation of meibomian glands (MGs), using limited annotations to reduce the workload on ophthalmologists and enhance the efficiency of clinical diagnosis. Totally 203 infrared meibomian images from 138 patients with dry eye disease, accompanied by corresponding annotations, were gathered for the study. A rectified scribble-supervised gland segmentation (RSSGS) model, incorporating temporal ensemble prediction, uncertainty estimation, and a transformation equivariance constraint, was introduced to address constraints imposed by limited supervision information inherent in scribble annotations. The viability and efficacy of the proposed model were assessed based on accuracy, intersection over union (IoU), and dice coefficient. Using manual labels as the gold standard, RSSGS demonstrated outcomes with an accuracy of 93.54%, a dice coefficient of 78.02%, and an IoU of 64.18%. Notably, these performance metrics exceed the current weakly supervised state-of-the-art methods by 0.76%, 2.06%, and 2.69%, respectively. Furthermore, despite achieving a substantial 80% reduction in annotation costs, it only lags behind fully annotated methods by 0.72%, 1.51%, and 2.04%. An innovative automatic segmentation model is developed for MGs in infrared eyelid images, using scribble annotation for training. This model maintains an exceptionally high level of segmentation accuracy while substantially reducing training costs. It holds substantial utility for calculating clinical parameters, thereby greatly enhancing the diagnostic efficiency of ophthalmologists in evaluating meibomian gland dysfunction.

Cluster2Former: Semisupervised Clustering Transformers for Video Instance Segmentation

A novel approach for video instance segmentation is presented using semisupervised learning. Our Cluster2Former model leverages scribble-based annotations for training, significantly reducing the need for comprehensive pixel-level masks. We augment a video instance segmenter, for example, the Mask2Former architecture, with similarity-based constraint loss to handle partial annotations efficiently. We demonstrate that despite using lightweight annotations (using only 0.5% of the annotated pixels), Cluster2Former achieves competitive performance on standard benchmarks. The approach offers a cost-effective and computationally efficient solution for video instance segmentation, especially in scenarios with limited annotation resources.

MicrobeMASST: a taxonomically informed mass spectrometry search tool for microbial metabolomics data

microbeMASST, a taxonomically informed mass spectrometry (MS) search tool, tackles limited microbial metabolite annotation in untargeted metabolomics experiments. Leveraging a curated database of >60,000 microbial monocultures, users can search known and unknown MS/MS spectra and link them to their respective microbial producers via MS/MS fragmentation patterns. Identification of microbe-derived metabolites and relative producers without a priori knowledge will vastly enhance the understanding of microorganisms’ role in ecology and human health.

Label-Efficient Point Cloud Semantic Segmentation: A Holistic Active Learning Approach

Deep learning models are the state of the art for semantic segmentation of point clouds, the success of which relies on the availability of large-scale annotated datasets. However, it can be prohibitively costly to prepare such datasets. In this work, we propose a holistic active learning (AL) approach to maximize model performance given limited annotation budgets. We investigate the appropriate sample granularity for active selection under the realistic “click” measurement of annotation cost, and demonstrate that superpoint-based selection allows for most efficient usage of the limited budget, when compared with point-level, polygon-level and instance/shape-level selection. We further propose new objective for AL acquisition function and exploit local consistency constraints to boost the performance of our superpoint-based approach. We evaluate our methods on three benchmark datasets, ShapeNet and PartNet and S3DIS. The results demonstrate that AL is an effective strategy to address the high annotation costs in semantic point cloud segmentation.

Automated Semantic Annotation Deploying Machine Learning Approaches: A Systematic Review

Semantic Web is the vision to make Internet data machine-readable to achieve information retrieval with higher granularity and personalisation. Semantic annotation is the process that binds machine-understandable descriptions into Web resources such as text and images. Hence, the success of Semantic Web dependson the wide availability of semantically annotated Web resources. However, there remains a huge amount of unannotated Web resources due to the limited annotation capability available. In order to address this, machine learning approaches have been used to improve the automation process. This Systematic Review aims to summarise the existing state-of-the-art literature to answer five Research Questions focusing on machine learning driven semantic annotation automation. The analysis of 40 selected primary studies reveals that the use of unitary and combination of machine learning algorithms are both the current directions. SupportVector Machine (SVM) is the most-used algorithm, and supervised learning is the predominant machine learning type. Both semi-automated and fully automated annotation are almost nearly achieved. Meanwhile, text is the most annotated Web resource; and the availability of third-party annotation tools is in-line with this. While Precision, Recall, F-Measure and Accuracy are the most deployed quality metrics, not all the studies measured the quality of the annotated results. In the future, standardising quality measures is the direction for research.

Uncertainty-guided cross learning via CNN and transformer for semi-supervised honeycomb lung lesion segmentation

Objective. Deep learning networks such as convolutional neural networks (CNN) and Transformer have shown excellent performance on the task of medical image segmentation, however, the usual problem with medical images is the lack of large-scale, high-quality pixel-level annotations, which is a very time-consuming and laborious task, and its further leads to compromised the performance of medical image segmentation under limited annotation conditions. Approach. In this paper, we propose a new semi-supervised learning method, uncertainty-guided cross learning, which uses a limited number of annotated samples along with a large number of unlabeled images to train the network. Specifically, we use two networks with different learning paradigms, CNN and Transformer, for cross learning, and use the prediction of one of them as a pseudo label to supervise the other, so that they can learn from each other, fully extract the local and global features of the images, and combine explicit and implicit consistency regularization constraints with pseudo label methods. On the other hand, we use epistemic uncertainty as a guiding message to encourage the model to learn high-certainty pixel information in high-confidence regions, and minimize the impact of erroneous pseudo labels on the overall learning process to improve the performance of semi-supervised segmentation methods. Main results. We conducted honeycomb lung lesion segmentation experiments using a honeycomb lung CT image dataset, and designed several sets of comparison experiments and ablation experiments to validate the effectiveness of our method. The final experimental results show that the Dice coefficient of our proposed method reaches 88.49% on the test set, and our method achieves state-of-the-art performance in honeycomb lung lesion segmentation compared to other semi-supervised learning methods. Significance. Our proposed method can effectively improve the accuracy of segmentation of honeycomb lung lesions, which provides an important reference for physicians in the diagnosis and treatment of this disease.

Reducing annotation burden in MR: A novel MR-contrast guided contrastive learning approach for image segmentation.

Contrastive learning, a successful form of representational learning, has shown promising results in pretraining deep learning (DL) models for downstream tasks. When working with limited annotation data, as in medical image segmentation tasks, learning domain-specific local representations can further improve the performance of DL models. In this work, we extend the contrastive learning framework to utilize domain-specific contrast information from unlabeled Magnetic Resonance (MR) images to improve the performance of downstream MR image segmentation tasks in the presence of limited labeled data. The contrast in MR images is controlled by underlying tissue properties (e.g., T1 or T2) and image acquisition parameters. We hypothesize that learning to discriminate local representations based on underlying tissue properties should improve subsequent segmentation tasks on MR images. We propose a novel constrained contrastive learning (CCL) strategy that uses tissue-specific information via a constraint map to define positive and negative local neighborhoods for contrastive learning, embedding this information in the representational space during pretraining. For a given MR contrast image, the proposed strategy uses local signal characteristics (constraint map) across a set of related multi-contrast MR images as a surrogate for underlying tissue information. We demonstrate the utility of the approach for downstream: (1) multi-organ segmentation tasks in T2-weighted images where a DL model learns T2 information with constraint maps from a set of 2D multi-echo T2-weighted images (n=101) and (2) tumor segmentation tasks in multi-parametric images from the public brain tumor segmentation (BraTS) (n=80) dataset where DL models learn T1 and T2 information from multi-parametric BraTS images. Performance is evaluated on downstream multi-label segmentation tasks with limited data in (1) T2-weighted images of the abdomen from an in-house Radial-T2 (Train/Test=30/20), (2) public Cartesian-T2 (Train/Test=6/12) dataset, and (3) multi-parametric MR images from the public brain tumor segmentation dataset (BraTS) (Train/Test=40/50). The performance of the proposed CCL strategy is compared to state-of-the-art self-supervised contrastive learning techniques. In each task, a model is also trained using all available labeled data for supervised baseline performance. The proposed CCL strategy consistently yielded improved Dice scores, Precision, and Recall metrics, and reduced HD95 values across all segmentation tasks. We also observed performance comparable to the baseline with reduced annotation effort. The t-SNE visualization of features for T2-weighted images demonstrates its ability to embed T2 information in the representational space. On the BraTS dataset, we also observed that using an appropriate multi-contrast space to learn T1+T2, T1, or T2 information during pretraining further improved the performance of tumor segmentation tasks. Learning to embed tissue-specific information that controls MR image contrast with the proposed constrained contrastive learning improved the performance of DL models on subsequent segmentation tasks compared to conventional self-supervised contrastive learning techniques. The use of such domain-specific local representations could help understand, improve performance, and mitigate the scarcity of labeled data in MR image segmentation tasks.

Progressive reinforcement learning for video summarization

Video summarization addresses generating video summaries to help watchers grasp the content of a video without watching it entirely. Many methods have engaged in automatic video summarization. Although these methods have performed well, they still suffer from limited training data and sparse reward problems. We propose a Progressive Reinforcement Learning Video Summarization structure (PRLVS) with an unsupervised reward. The reward measures the information and quality the selected frames convey without annotations. Striving to earn higher rewards, our PRLVS adopts a “T”-type human thinking paradigm: choosing some key frames and checking if their adjacent frames are better than them. To simulate this paradigm, we decompose the flat strategy into a hierarchical strategy consisting of a horizontal policy and a vertical policy. These two policies are optimized alternatively, which densifies the reward while reducing the exploration space. Their cooperation also makes the agent capture the context information of the whole video at every step. Extensive experimental results on two benchmark databases (i.e., SumMe, TVSum) show that our PRLVS outperforms the comparisons and approaches the supervised methods, which indicates that it is significant to integrate our unsupervised reward into the progressive reinforcement learning structure to address limited annotation and sparse reward problems.

Lithological mapping of geological remote sensing via adversarial semi-supervised segmentation network

Geological remote sensing interpretation (GRSI), which aims to recognize multiple geological elements based on their characteristics on satellite remote sensing images, is vital in large-scale regional lithological mapping. However, due to the influence of long-term geological movements, the spatial distribution of geological elements (such as lithology, glaciers, and soils) on the image is often complex and highly fragmented. In addition, the characteristics of high inter-class similarity and severe homogenization make the annotation of geological element samples require significant cost and expert knowledge. These lead to insufficient interpretation accuracy and limited annotation samples in GRSI. To alleviate the dependence of labeled samples and promote the performance of GRSI, we propose an adversarial semi-supervised segmentation network with object-context and global-attention (AdvSemi-OCGNet) for the GRSI, which achieves effective segmentation results in the case of limited labeled samples. Under the architecture of adversarial learning, a proposed baseline network OCGNet and a full convolution discriminator (FCD) are integrated to conduct semi-supervised segmentation. OCGNet, as the generator, aims to confuse FCD by predicting probability maps. Then FCD selects trustworthy regions with high-confidence of unlabeled sample to generate pseudo-labels for semi-supervised segmentation of OCGNet. Iterative adversarial learning is employed to simulate the process of expert interpretation of geological elements through continuous discrimination and correction. Finally, a fully connected conditional random field eliminates holes and isolated areas of segmented results caused by misclassification. Two study areas are selected, which include various geological elements such as multiple lithologies, soils, rock glaciers, and surface water. Numerous experiments have revealed the superiority of the proposed model in GRSI with limited annotation samples.