Large Intra-class Variations Research Articles

Multi-view visual semantic embedding for cross-modal image–text retrieval

Visual Semantic Embedding (VSE) is a dominant method for cross-modal image–text retrieval. The purpose of VSE is to learn an embedding space where images can be embedded close to the corresponding captions. However, there are large intra-class variations in image–text data. Multiple captions describing the same image may be described from different views, and descriptions of different views are often dissimilar. The VSE method embeds samples from the same class in similar positions, which suppresses intra-class variations and leads to inferior generalization. This paper proposes a Multi-View Visual Semantic Embedding (MV-VSE) framework that learns multiple embeddings for an image, explicitly modeling intra-class variation. To optimize the MV-VSE framework, a multi-view triplet loss is proposed, which jointly optimizes multi-view embeddings while retaining intra-class variation. Recently, large-scale Vision-Language Pre-training (VLP) has become a new paradigm for cross-modal image–text retrieval. To allow our framework to be flexibly applied to the traditional VSE models and VSE-based VLP models, we incorporate the contrastive loss commonly used in VLP and the triplet loss into a unified loss, and further propose a multi-view unified loss. Our framework can be applied plug-and-play to traditional VSE models and VSE-based VLP models without excessively increasing model complexity. Experimental results on the image–text retrieval benchmark datasets demonstrate that applying our framework can boost the retrieval performance of current VSE models. The code is available at https://github.com/AAA-Zheng/MV-VSE.

KBCNet: Independently keypoint learning for Small Object Semantic Correspondence

Establishing semantic correspondence between pairs of images within the same category or similar scenes is challenging due to large intra-class variations. In this paper, we introduce a novel problem termed ’Small Object Semantic Correspondence (SOSC).’ The challenge in SOSC arises from the close proximity of keypoints in small objects, leading to the fusion of their respective features, which complicates accurate keypoint prediction and correspondence identification. To address this issue, we propose the Keypoint Bounding box-centered Cropping (KBC) method, designed to increase the spatial separation between keypoints in small objects, thereby enabling the independent learning of these keypoints. The KBC method is seamlessly integrated into our proposed inference pipeline and can be easily incorporated into other methodologies, resulting in significant performance improvements. Furthermore, we introduce a novel framework called KBCNet, which serves as our baseline model. KBCNet includes a Cross-Scale Feature Alignment (CSFA) module and an efficient 4D convolutional decoder. The CSFA module aligns multi-scale features, enriching keypoint representations by integrating fine-grained details with deep semantic features. Meanwhile, the 4D convolutional decoder, based on efficient 4D convolution, ensures both high efficiency and rapid convergence. To empirically validate the effectiveness of our approach, we conducted extensive experiments on three widely used benchmarks: PF-PASCAL, PF-WILLOW, and SPair-71k. Our KBC method demonstrated a substantial performance improvement of 7.5% on the SPair-71k dataset, providing strong evidence of its efficacy.

A lightweight dual-attention network for tomato leaf disease identification.

Tomato disease image recognition plays a crucial role in agricultural production. Today, while machine vision methods based on deep learning have achieved some success in disease recognition, they still face several challenges. These include issues such as imbalanced datasets, unclear disease features, small inter-class differences, and large intra-class variations. To address these challenges, this paper proposes a method for classifying and recognizing tomato leaf diseases based on machine vision. First, to enhance the disease feature details in images, a piecewise linear transformation method is used for image enhancement, and oversampling is employed to expand the dataset, compensating for the imbalanced dataset. Next, this paper introduces a convolutional block with a dual attention mechanism called DAC Block, which is used to construct a lightweight model named LDAMNet. The DAC Block innovatively uses Hybrid Channel Attention (HCA) and Coordinate Attention (CSA) to process channel information and spatial information of input images respectively, enhancing the model's feature extraction capabilities. Additionally, this paper proposes a Robust Cross-Entropy (RCE) loss function that is robust to noisy labels, aimed at reducing the impact of noisy labels on the LDAMNet model during training. Experimental results show that this method achieves an average recognition accuracy of 98.71% on the tomato disease dataset, effectively retaining disease information in images and capturing disease areas. Furthermore, the method also demonstrates strong recognition capabilities on rice crop disease datasets, indicating good generalization performance and the ability to function effectively in disease recognition across different crops. The research findings of this paper provide new ideas and methods for the field of crop disease recognition. However, future research needs to further optimize the model's structure andcomputational efficiency, and validate its application effects in more practical scenarios.

SPD Manifold Deep Metric Learning for Image Set Classification.

By characterizing each image set as a nonsingular covariance matrix on the symmetric positive definite (SPD) manifold, the approaches of visual content classification with image sets have made impressive progress. However, the key challenge of unhelpfully large intraclass variability and interclass similarity of representations remains open to date. Although, several recent studies have mitigated the two problems by jointly learning the embedding mapping and the similarity metric on the original SPD manifold, their inherent shallow and linear feature transformation mechanism are not powerful enough to capture useful geometric features, especially in complex scenarios. To this end, this article explores a novel approach, termed SPD manifold deep metric learning (SMDML), for image set classification. Specifically, SMDML first selects a prevailing SPD manifold neural network (SPDNet) as the backbone (encoder) to derive an SPD matrix nonlinear representation. To counteract the degradation of structural information during multistage feature embedding, we construct a Riemannian decoder at the end of the encoder, trained by a reconstruction error term (RT), to induce the generated low-dimensional feature manifold of the hidden layer to capture the pivotal information about the visual data describing the imaged scene. We demonstrate through theory and experiments that it is feasible to replace the Riemannian metric with Euclidean distance in RT. Then, the ReCov layer is introduced into the established Riemannian network to regularize the local statistical information within each input feature matrix, which enhances the effectiveness of the learning process. The theoretical analysis of the activation function used in the ReCov layer in terms of continuity and conditions for generating positive definite matrices is beneficial for network design. Inspired by the fact that the single cross-entropy loss used for training is unable to effectively parse the geometric distribution of the deep representations, we finally endow the suggested model with a novel metric learning regularization term. By explicitly incorporating the encoding and processing of the data variations into the network learning process, this term can not only derive a powerful Riemannian representation but also train an effective classifier. The experimental results show the superiority of the proposed approach on three typical visual classification tasks.

MIST: Multi-instance selective transformer for histopathological subtype prediction

Accurate histopathological subtype prediction is clinically significant for cancer diagnosis and tumor microenvironment analysis. However, achieving accurate histopathological subtype prediction is a challenging task due to (1) instance-level discrimination of histopathological images, (2) low inter-class and large intra-class variances among histopathological images in their shape and chromatin texture, and (3) heterogeneous feature distribution over different images. In this paper, we formulate subtype prediction as fine-grained representation learning and propose a novel multi-instance selective transformer (MIST) framework, effectively achieving accurate histopathological subtype prediction. The proposed MIST designs an effective selective self-attention mechanism with multi-instance learning (MIL) and vision transformer (ViT) to adaptive identify informative instances for fine-grained representation. Innovatively, the MIST entrusts each instance with different contributions to the bag representation based on its interactions with instances and bags. Specifically, a SiT module with selective multi-head self-attention (S-MSA) is well-designed to identify the representative instances by modeling the instance-to-instance interactions. On the contrary, a MIFD module with the information bottleneck is proposed to learn the discriminative fine-grained representation for histopathological images by modeling instance-to-bag interactions with the selected instances. Substantial experiments on five clinical benchmarks demonstrate that the MIST achieves accurate histopathological subtype prediction and obtains state-of-the-art performance with an accuracy of 0.936. The MIST shows great potential to handle fine-grained medical image analysis, such as histopathological subtype prediction in clinical applications.

Hierarchical online contrastive anomaly detection for fetal arrhythmia diagnosis in ultrasound

Arrhythmia is a major cardiac abnormality in fetuses. Therefore, early diagnosis of arrhythmia is clinically crucial. Pulsed-wave Doppler ultrasound is a commonly used diagnostic tool for fetal arrhythmia. Its key step for diagnosis involves identifying adjacent measurable cardiac cycles (MCCs). As cardiac activity is complex and the experience of sonographers is often varied, automation can improve user-independence and diagnostic-validity. However, arrhythmias pose several challenges for automation because of complex waveform variations, which can cause major localization bias and missed or false detection of MCCs. Filtering out non-MCC anomalies is difficult because of large intra-class and small inter-class variations between MCCs and non-MCCs caused by agnostic morphological waveform variations. Moreover, rare arrhythmia cases are insufficient for classification algorithms to adequately learn discriminative features. Using only normal cases for training, we propose a novel hierarchical online contrastive anomaly detection (HOCAD) framework for arrhythmia diagnosis during test time. The contribution of this study is three-fold. First, we develop a coarse-to-fine framework inspired by hierarchical diagnostic logic, which can refine localization and avoid missed detection of MCCs. Second, we propose an online learning-based contrastive anomaly detection with two new anomaly scores, which can adaptively filter out non-MCC anomalies on a single image during testing. With these complementary efforts, we precisely determine MCCs for correct measurements and diagnosis. Third, to the best of our knowledge, this is the first reported study investigating intelligent diagnosis of fetal arrhythmia on a large-scale and multi-center ultrasound dataset. Extensive experiments on 3850 cases, including 266 cases covering three typical types of arrhythmias, demonstrate the effectiveness of the proposed framework.

Graph Representation and Prototype Learning for webly supervised fine-grained image recognition

Webly supervised fine-grained image recognition (FGIR) learns to distinguish sub-ordinate categories based on webly-retrieved data, which can dramatically alleviate the dependency on manually annotated labels. This is quite a challenging task due to the heavy noise labels and the inherent dilemma of small inter-class variance and large intra-class variance. Current webly supervised algorithms learn holistic category prototypes to help correct noisy labels but ignore local features that can distinguish different sub-ordinate categories. In this work, we propose a Graph Representation and Prototype Learning (GRPL) framework to automatically mine discriminative local regions and their interactions with holistic image to learn instance graph representation both category graph prototype to help correct noisy labels and retrieve out-of-distribution (OOD) samples. Specifically, an attention-focused module is designed to extract the discriminative regions and then build a structured graph to correlate them with the holistic image for each instance and an identical graph to model holistic-local correlations for each category. Next, we apply two stacked graph convolution networks to explore holistic-local interaction within each graph and across two graphs to learn graph representation for each instance and graph prototype for each category. Finally, the similarities between the instance-level and prototype-level graph representation are learned to help correct noisy labels and exclude OOD samples. Extensive experiments conducted on several datasets show the proposed approach achieves superior performance compared with current leading algorithms.

A class distribution learning method for few-shot remote sensing scene classification

ABSTRACT In recent years, few-shot remote sensing scene classification has become an important foundational research topic, aiming to achieve better classification performance with limited labeled data. Due to the large intra-class variances and inter-class similarity of remote sensing scenes, the scene classification task for remote sensing is much more challenging than general few-shot image classification. In this letter, we propose a class distribution learning method under limited samples by fully using the support samples, which consists of intra-class feature aggregation (IFA) module, intra-class feature homogenization module (IFH) and inter-class cross calibration (ICC) module. IFA and IFH modules are proposed to enable the aggregation and even distribution of intra-class features around the prototypes generated by the model, reducing intra-class distances and mitigating the generation of extreme samples near decision boundaries. ICC module is designed to encourage the features generated by the model for inter-class to be further apart, thereby increasing the inter-class distance and reducing the probability of classification errors. To demonstrate the effectiveness of our proposed method, we conduct comprehensive experiments on three benchmark remote sensing scene datasets: NWPU-RESISC45, WHU-RS19, and UC Merced, achieving extremely advanced performance. Our source codes have been available at https://github.com/mitsuha97/CDLM.

An efficient Fusion-Purification Network for Cervical pap-smear image classification

Background and Objectives:In cervical cell diagnostics, autonomous screening technology constitutes the foundation of automated diagnostic systems. Currently, numerous deep learning-based classification techniques have been successfully implemented in the analysis of cervical cell images, yielding favorable outcomes. Nevertheless, efficient discrimination of cervical cells continues to be challenging due to large intra-class and small inter-class variations. The key to dealing with this problem is to capture localized informative differences from cervical cell images and to represent discriminative features efficiently. Existing methods neglect the importance of global morphological information, resulting in inadequate feature representation capability. Methods:To address this limitation, we propose a novel cervical cell classification model that focuses on purified fusion information. Specifically, we first integrate the detailed texture information and morphological structure features, named cervical pathology information fusion. Second, in order to enhance the discrimination of cervical cell features and address the data redundancy and bias inherent after fusion, we design a cervical purification bottleneck module. This model strikes a balance between leveraging purified features and facilitating high-efficiency discrimination. Furthermore, we intend to unveil a more intricate cervical cell dataset: Cervical Cytopathology Image Dataset (CCID). Results:Extensive experiments on two real-world datasets show that our proposed model outperforms state-of-the-art cervical cell classification models. Conclusions:The results show that our method can well help pathologists to accurately evaluate cervical smears.

Polymorphic Clustering and Approximate Masking Framework for Fine-Grained Insect Image Classification

Insect diversity monitoring is crucial for biological pest control in agriculture and forestry. Modern monitoring of insect species relies heavily on fine-grained image classification models. Fine-grained image classification faces challenges such as small inter-class differences and large intra-class variances, which are even more pronounced in insect scenes where insect species often exhibit significant morphological differences across multiple life stages. To address these challenges, we introduce segmentation and clustering operations into the image classification task and design a novel network model training framework for fine-grained classification of insect images using multi-modality clustering and approximate mask methods, named PCAM-Frame. In the first stage of the framework, we adopt the Polymorphic Clustering Module, and segmentation and clustering operations are employed to distinguish various morphologies of insects at different life stages, allowing the model to differentiate between samples at different life stages during training. The second stage consists of a feature extraction network, called Basenet, which can be any mainstream network that performs well in fine-grained image classification tasks, aiming to provide pre-classification confidence for the next stage. In the third stage, we apply the Approximate Masking Module to mask the common attention regions of the most likely classes and continuously adjust the convergence direction of the model during training using a Deviation Loss function. We apply PCAM-Frame with multiple classification networks as the Basenet in the second stage and conduct extensive experiments on the Insecta dataset of iNaturalist 2017 and IP102 dataset, achieving improvements of 2.2% and 1.4%, respectively. Generalization experiments on other fine-grained image classification datasets such as CUB200-2011 and Stanford Dogs also demonstrate positive effects. These experiments validate the pertinence and effectiveness of our framework PCAM-Frame in fine-grained image classification tasks under complex conditions, particularly in insect scenes.

Few-Shot Class-Incremental Learning for Medical Time Series Classification.

Continuously analyzing medical time series as new classes emerge is meaningful for health monitoring and medical decision-making. Few-shot class-incremental learning (FSCIL) explores the classification of few-shot new classes without forgetting old classes. However, little of the existing research on FSCIL focuses on medical time series classification, which is more challenging to learn due to its large intra-class variability. In this paper, we propose a framework, the Meta self-Attention Prototype Incrementer (MAPIC) to address these problems. MAPIC contains three main modules: an embedding encoder for feature extraction, a prototype enhancement module for increasing inter-class variation, and a distance-based classifier for reducing intra-class variation. To mitigate catastrophic forgetting, MAPIC adopts a parameter protection strategy in which the parameters of the embedding encoder module are frozen at incremental stages after being trained in the base stage. The prototype enhancement module is proposed to enhance the expressiveness of prototypes by perceiving inter-class relations using a self-attention mechanism. We design a composite loss function containing the sample classification loss, the prototype non-overlapping loss, and the knowledge distillation loss, which work together to reduce intra-class variations and resist catastrophic forgetting. Experimental results on three different time series datasets show that MAPIC significantly outperforms state-of-the-art approaches by 27.99%, 18.4%, and 3.95%, respectively.

Delving into Multimodal Prompting for Fine-Grained Visual Classification

Fine-grained visual classification (FGVC) involves categorizing fine subdivisions within a broader category, which poses challenges due to subtle inter-class discrepancies and large intra-class variations. However, prevailing approaches primarily focus on uni-modal visual concepts. Recent advancements in pre-trained vision-language models have demonstrated remarkable performance in various high-level vision tasks, yet the applicability of such models to FGVC tasks remains uncertain. In this paper, we aim to fully exploit the capabilities of cross-modal description to tackle FGVC tasks and propose a novel multimodal prompting solution, denoted as MP-FGVC, based on the contrastive language-image pertaining (CLIP) model. Our MP-FGVC comprises a multimodal prompts scheme and a multimodal adaptation scheme. The former includes Subcategory-specific Vision Prompt (SsVP) and Discrepancy-aware Text Prompt (DaTP), which explicitly highlights the subcategory-specific discrepancies from the perspectives of both vision and language. The latter aligns the vision and text prompting elements in a common semantic space, facilitating cross-modal collaborative reasoning through a Vision-Language Fusion Module (VLFM) for further improvement on FGVC. Moreover, we tailor a two-stage optimization strategy for MP-FGVC to fully leverage the pre-trained CLIP model and expedite efficient adaptation for FGVC. Extensive experiments conducted on four FGVC datasets demonstrate the effectiveness of our MP-FGVC.

Multi-Prototype Space Learning for Commonsense-Based Scene Graph Generation

In the domain of scene graph generation, modeling commonsense as a single-prototype representation has been typically employed to facilitate the recognition of infrequent predicates. However, a fundamental challenge lies in the large intra-class variations of the visual appearance of predicates, resulting in subclasses within a predicate class. Such a challenge typically leads to the problem of misclassifying diverse predicates due to the rough predicate space clustering. In this paper, inspired by cognitive science, we maintain multi-prototype representations for each predicate class, which can accurately find the multiple class centers of the predicate space. Technically, we propose a novel multi-prototype learning framework consisting of three main steps: prototype-predicate matching, prototype updating, and prototype space optimization. We first design a triple-level optimal transport to match each predicate feature within the same class to a specific prototype. In addition, the prototypes are updated using momentum updating to find the class centers according to the matching results. Finally, we enhance the inter-class separability of the prototype space through iterations of the inter-class separability loss and intra-class compactness loss. Extensive evaluations demonstrate that our approach significantly outperforms state-of-the-art methods on the Visual Genome dataset.

Hyperspectral image classification using Walsh Hadamard transform- based key band selection and deep convolutional neural networks

ABSTRACT Hyperspectral images (HSIs) have transformed the field of remote sensing by providing researchers with a wealth of information about the Earth’s surface. However, analyzing these images can be an overwhelming task due to the presence of overlapping areas, nested regions, and large intra-class variability. Hyperspectral image classification (HSIC) is a crucial part of identifying the various land cover classes present in hyperspectral images. In order to enhance the accuracy of HSIC, researchers utilize the potential of three-dimensional convolutional neural networks (3D-CNNs). With the ability to influence both the spectral and spatial data present in HSIs, 3D-CNNs provide a promising solution to overcome the challenges associated with HSIC. In this paper, a new method for key band selection is proposed to improve the performance of 3D-CNN model. The proposed method selects the most relevant key bands based on Walsh-Hadamard kernel strength features. These key bands are then used to extract overlapping 3D spatial patches, which serve as input to the proposed 3D-CNN model. To evaluate the performance of the 3D-CNN model six standard benchmark datasets are used. The effectiveness of the proposed method improves the performance of 3D-CNN for HSIC.

Category attention guided network for semantic segmentation of Fine-Resolution remote sensing images

The semantic segmentation task is an essential issue in various fields, including land cover classification and cultural heritage investigation. The CNN and Transformer have been widely utilized in semantic segmentation tasks due to notable advancements in deep learning technologies. However, these methodologies may not fully account for remote sensing images' distinctive attributes, including the large intra-class variation and the small inter-class variation. Driven by it, we propose a category attention guided network (CAGNet). Initially, a local feature extraction module is devised to cater to striped objects and features at different scales. Then, we propose a novel concept of category attention for remote sensing images as a feature representation of category differences between pixels. Meanwhile, we designed the Transformer-based and CNN-based category attention guided modules to integrate the proposed category attention into the global scoring functions and local category feature weights, respectively. The network is designed to give more attention to the category features by updating these weights during the training process. Finally, a feature fusion module is developed to integrate global, local, and category multi-scale features and contextual information. A series of extensive experiments along with ablation studies on the UAVid, Vaihingen, and Potsdam datasets indicate that our network outperforms existing methods, including those based on CNN and Transformer.

FPIRST: Fatigue Driving Recognition Method Based on Feature Parameter Images and a Residual Swin Transformer.

Fatigue driving is a serious threat to road safety, which is why accurately identifying fatigue driving behavior and warning drivers in time are of great significance in improving traffic safety. However, accurately recognizing fatigue driving is still challenging due to large intra-class variations in facial expression, continuity of behaviors, and illumination conditions. A fatigue driving recognition method based on feature parameter images and a residual Swin Transformer is proposed in this paper. First, the face region is detected through spatial pyramid pooling and a multi-scale feature output module. Then, a multi-scale facial landmark detector is used to locate 23 key points on the face. The aspect ratios of the eyes and mouth are calculated based on the coordinates of these key points, and a feature parameter matrix for fatigue driving recognition is obtained. Finally, the feature parameter matrix is converted into an image, and the residual Swin Transformer network is presented to recognize fatigue driving. Experimental results on the HNUFD dataset show that the proposed method achieves an accuracy of 96.512%, thus outperforming state-of-the-art methods.

Prototype Correlation Matching and Class-Relation Reasoning for Few-Shot Medical Image Segmentation.

Few-shot medical image segmentation has achieved great progress in improving accuracy and efficiency of medical analysis in the biomedical imaging field. However, most existing methods cannot explore inter-class relations among base and novel medical classes to reason unseen novel classes. Moreover, the same kind of medical class has large intra-class variations brought by diverse appearances, shapes and scales, thus causing ambiguous visual characterization to degrade generalization performance of these existing methods on unseen novel classes. To address the above challenges, in this paper, we propose a Prototype correlation Matching and Class-relation Reasoning (i.e., PMCR) model. The proposed model can effectively mitigate false pixel correlation matches caused by large intra-class variations while reasoning inter-class relations among different medical classes. Specifically, in order to address false pixel correlation match brought by large intra-class variations, we propose a prototype correlation matching module to mine representative prototypes that can characterize diverse visual information of different appearances well. We aim to explore prototypelevel rather than pixel-level correlation matching between support and query features via optimal transport algorithm to tackle false matches caused by intra-class variations. Meanwhile, in order to explore inter-class relations, we design a class-relation reasoning module to segment unseen novel medical objects via reasoning inter-class relations between base and novel classes. Such inter-class relations can be well propagated to semantic encoding of local query features to improve few-shot segmentation performance. Quantitative comparisons illustrates the large performance improvement of our model over other baseline methods.

Dual-branch Branch Networks Based on Contrastive Learning for Long-Tailed Remote Sensing

Deep learning has been widely used in remote sensing image classification and achieves many excellent results. These methods are all based on relatively balanced data sets. However, in real-world scenarios, many data sets belong to the long-tailed distribution, resulting in poor performance. In view of the good performance of contrastive learning in long-tailed image classification, a new dual-branch fusion learning classification model is proposed to fuse the discriminative features of remote sensing images with spatial data, making full use of valuable image representation information in imbalance data. This paper also presents a hybrid loss, which solves the problem of poor discrimination of extracted features caused by large intra-class variation and inter-class ambiguity. Extended experiments on three long-tailed remote sensing image classification data sets demonstrate the advantages of the proposed dual-branch model based on contrastive learning in long-tailed image classification.

Fine-grained Recognition with Learnable Semantic Data Augmentation.

Fine-grained image recognition is a longstanding computer vision challenge that focuses on differentiating objects belonging to multiple subordinate categories within the same meta-category. Since images belonging to the same meta-category usually share similar visual appearances, mining discriminative visual cues is the key to distinguishing fine-grained categories. Although commonly used image-level data augmentation techniques have achieved great success in generic image classification problems, they are rarely applied in fine-grained scenarios, because their random editing-region behavior is prone to destroy the discriminative visual cues residing in the subtle regions. In this paper, we propose diversifying the training data at the feature-level to alleviate the discriminative region loss problem. Specifically, we produce diversified augmented samples by translating image features along semantically meaningful directions. The semantic directions are estimated with a covariance prediction network, which predicts a sample-wise covariance matrix to adapt to the large intra-class variation inherent in fine-grained images. Furthermore, the covariance prediction network is jointly optimized with the classification network in a meta-learning manner to alleviate the degenerate solution problem. Experiments on four competitive fine-grained recognition benchmarks (CUB-200-2011, Stanford Cars, FGVC Aircrafts, NABirds) demonstrate that our method significantly improves the generalization performance on several popular classification networks (e.g., ResNets, DenseNets, EfficientNets, RegNets and ViT). Combined with a recently proposed method, our semantic data augmentation approach achieves state-of-the-art performance on the CUB-200-2011 dataset. The source code will be released.

Multi-Surface Multi-Technique (MUST) Latent Fingerprint Database

Latent fingerprint recognition involves acquisition and comparison of latent fingerprints with an exemplar gallery of fingerprints. The diversity in the type of surface leads to different procedures to recover the latent fingerprint. The appearance of latent fingerprints vary significantly due to the development techniques, leading to large intra-class variation. Due to lack of large datasets acquired using multiple mechanisms and surfaces, existing algorithms for latent fingerprints enhancement and comparison may perform poorly. In this study, we propose a Multi-Surface Multi-Technique (MUST) Latent Fingerprint Database. The database consists of more than 16,000 latent fingerprint impressions from 120 unique classes (120 fingers from 12 participants). Including corresponding exemplar fingerprints (livescan and rolled) and extended gallery, the dataset has nearly 21,000 impressions. It has latent fingerprints acquired under 35 different scenarios and additional four subsets of exemplar prints captured using live scan sensor and inked-rolled prints. With 39 different subsets, the database illustrates intra-class variations in latent fingerprints. The database has a potential usage towards building robust algorithms for latent fingerprint enhancement, segmentation, comparison, and multi-task learning. We also provide annotations for manually marked minutiae, acquisition PPI, and semantic segmentation masks are also provided. We also present the experimental protocol and the baseline results for the proposed dataset. We assert that the availability of the proposed database can encourage research in handling intra-class variation in latent fingerprint recognition.