Cross-modal Retrieval Methods Research Articles

Semantic Pre-Alignment and Ranking Learning With Unified Framework for Cross-Modal Retrieval

Cross-modal retrieval aims at retrieving highly semantic relevant information among multi-modalities. Existing cross-modal retrieval methods mainly explore the semantic consistency between image and text while rarely consider the rankings of positive instances in the retrieval results. Moreover, these methods seldom take into account the cross-interaction between image and text, which leads to the deficiency of learning their semantic relations. In this paper, we propose a Unified framework with Ranking Learning (URL) for cross-modal retrieval. The unified framework consists of three sub-networks, visual network, textual network, and interaction network. Visual network and textual network project the image feature and text feature into their corresponding hidden spaces respectively. Then, the interaction network forces the target image-text representation to align in the common space. For unifying both semantics and rankings, we propose a new optimization paradigm including pre-alignment for semantic knowledge transfer and ranking learning for final retrieval, which can decouple semantic alignment and ranking learning. The former focuses on the semantic pre-alignment optimized by semantic classification and the latter revolves around the retrieval rankings. For the ranking learning, we introduce a cross-AP loss which can directly optimize the retrieval metric average precision for cross-modal retrieval. We conduct experiments on four widely-used benchmarks, including Wikipedia dataset, Pascal Sentence dataset, NUS-WIDE-10k dataset, and PKU XMediaNet dataset respectively. Extensive experimental results show that the proposed method can obtain higher retrieval precision.

Dual-Pathway Deep Hashing-Based Adversarial Learning for Cross-Modal Retrieval

The application of deep hashing methods for cross-modal retrieval has seen growing interest due to their storage efficiency and fast query execution. However, the challenge posed by the “heterogeneity gap” in multi-modal datasets cannot be understated. To address this, we present a novel framework named Dual-Pathway Deep Hashing-Based Adversarial Learning (DP-DHAL), engineered to surmount this challenge. The architecture of DP-DHAL integrates three key components: (a) a dual-pathway representation learning module tasked with extracting modality-specific features; (b) an adversarial module working to align the distributions of cross-modal features; and (c) a deep hashing module responsible for generating hash codes that uphold the similarity relationships across different modalities. Additionally, we have developed a unique Hamming triplet-margin loss function to refine the assessment of content similarities. The DP-DHAL model is trained through an adversarial process where the adversarial module’s goal is to discern cross-modal features with the aim of reducing the heterogeneity gap. Simultaneously, the representation learning module is focused on producing representations that can both deceive the adversarial module and preserve cross-modal similarities to yield distinctive hash codes. Comprehensive experiments on varied datasets have shown that our proposed method outperforms other leading cross-modal hashing techniques.

Cross-modal retrieval based on multi-dimensional feature fusion hashing

Along with the continuous breakthrough and popularization of information network technology, multi-modal data, including texts, images, videos, and audio, is growing rapidly. We can retrieve different modal data to meet our needs, so cross-modal retrieval has important theoretical significance and application value. In addition, because the data of different modalities can be mutually retrieved by mapping them to a unified Hamming space, hash codes have been extensively used in the cross-modal retrieval field. However, existing cross-modal hashing models generate hash codes based on single-dimension data features, ignoring the semantic correlation between data features in different dimensions. Therefore, an innovative cross-modal retrieval method using Multi-Dimensional Feature Fusion Hashing (MDFFH) is proposed. To better get the image’s multi-dimensional semantic features, a convolutional neural network, and Vision Transformer are combined to construct an image multi-dimensional fusion module. Similarly, we apply the multi-dimensional text fusion module to the text modality to obtain the text’s multi-dimensional semantic features. These two modules can effectively integrate the semantic features of data in different dimensions through feature fusion, making the generated hash code more representative and semantic. Extensive experiments and corresponding analysis results on two datasets indicate that MDFFH’s performance outdoes other baseline models.

Video and audio are images: A cross-modal mixer for original data on video–audio retrieval

Cross-modal retrieval has become popular in recent years, particularly with the rise of multimedia. Generally, the information from each modality exhibits distinct representations and semantic information, which makes feature tends to be in separate latent spaces encoded with dual-tower architecture and makes it challenging to establish semantic relationships between modalities, resulting in poor retrieval performance. To address this issue, we propose a novel framework for cross-modal retrieval consisting of a cross-modal mixer, a masked autoencoder for pre-training, and a cross-modal retriever for downstream tasks. Specifically, we first adopt cross-modal mixer and mask modeling to fuse the original modality and eliminate redundancy. Then, an encoder–decoder architecture is applied to achieve a fuse-then-separate task in the pre-training phase. We feed masked fused representations into the encoder and reconstruct them with the decoder, ultimately separating the original data of two modalities. We use the pre-trained encoder in downstream tasks to build the cross-modal retrieval method. Extensive experiments on 2 real-world datasets show that our approach outperforms previous state-of-the-art methods in video–audio matching tasks, improving retrieval accuracy by up to 2×. Furthermore, we prove our model performance by transferring it to other downstream tasks as a universal model.

Disambiguity and Alignment: An Effective Multi-Modal Alignment Method for Cross-Modal Recipe Retrieval.

As a prominent topic in food computing, cross-modal recipe retrieval has garnered substantial attention. However, the semantic alignment across food images and recipes cannot be further enhanced due to the lack of intra-modal alignment in existing solutions. Additionally, a critical issue named food image ambiguity is overlooked, which disrupts the convergence of models. To these ends, we propose a novel Multi-Modal Alignment Method for Cross-Modal Recipe Retrieval (MMACMR). To consider inter-modal and intra-modal alignment together, this method measures the ambiguous food image similarity under the guidance of their corresponding recipes. Additionally, we enhance recipe semantic representation learning by involving a cross-attention module between ingredients and instructions, which is effective in supporting food image similarity measurement. We conduct experiments on the challenging public dataset Recipe1M; as a result, our method outperforms several state-of-the-art methods in commonly used evaluation criteria.

Improving semantic video retrieval models by training with a relevance-aware online mining strategy

To retrieve a video via a multimedia search engine, a textual query is usually created by the user and then used to perform the search. Recent state-of-the-art cross-modal retrieval methods learn a joint text–video embedding space by using contrastive loss functions, which maximize the similarity of positive pairs while decreasing that of the negative pairs. Although the choice of these pairs is fundamental for the construction of the joint embedding space, the selection procedure is usually driven by the relationships found within the dataset: a positive pair is commonly formed by a video and its own caption, whereas unrelated video-caption pairs represent the negative ones. We hypothesize that this choice results in a retrieval system with limited semantics understanding, as the standard training procedure requires the system to discriminate between groundtruth and negative even though there is no difference in their semantics. Therefore, differently from the previous approaches, in this paper we propose a novel strategy for the selection of both positive and negative pairs which takes into account both the annotations and the semantic contents of the captions. By doing so, the selected negatives do not share semantic concepts with the positive pair anymore, and it is also possible to discover new positives within the dataset. Based on our hypothesis, we provide a novel design of two popular contrastive loss functions, and explore their effectiveness on four heterogeneous state-of-the-art approaches. The extensive experimental analysis conducted on four datasets, EPIC-Kitchens-100, MSR-VTT, MSVD, and Charades, validates the effectiveness of the proposed strategy, observing, e.g., more than +20% nDCG on EPIC-Kitchens-100. Furthermore, these results are corroborated with qualitative evidence both supporting our hypothesis and explaining why the proposed strategy effectively overcomes it.

Deep Neighborhood-aware Proxy Hashing with Uniform Distribution Constraint for Cross-modal Retrieval

Cross-modal retrieval methods based on hashing have gained significant attention in both academic and industrial research. Deep learning techniques have played a crucial role in advancing supervised cross-modal hashing methods, leading to significant practical improvements. Despite these achievements, current deep cross-modal hashing still encounters some underexplored limitations. Specifically, most of the available deep hashing usually utilizes pair-wise or triplet-wise strategies to promote the separation of the inter-classes by calculating the relative similarities between samples, weakening the compactness of intra-class data from different modalities, which could generate ambiguous neighborhoods. In this article, the Deep Neighborhood-aware Proxy Hashing (DNPH) framework is proposed to learn a discriminative embedding space with the original neighborhood relation preserved. By introducing learnable shared category proxies, the neighborhood-aware proxy loss is proposed to project the heterogeneous data into a unified common embedding, in which the sample is pulled closer to the corresponding category proxy and is pushed away from other proxies, capturing small within-class scatter and big between-class scatter. To enhance the quality of the obtained binary codes, the uniform distribution constraint is developed to make each hash bit independently obey the discrete uniform distribution. In addition, the discrimination loss is designed to preserve modality-specific semantic information of samples. Extensive experiments are performed on three benchmark datasets to prove that our proposed DNPH framework achieves comparable or even better performance compared with the state-of-the-art cross-modal retrieval applications. The corresponding code implementation of our DNPH framework is as follows: https://github.com/QinLab-WFU/OUR-DNPH .

Unpaired robust hashing with noisy labels for zero-shot cross-modal retrieval

With new social media concepts emerging, zero-shot cross-modal retrieval methods have gained significant attention. Most of the existing methods assume that the labels of training data are correct and the different modalities are perfectly matched, which is unrealistic in real-life retrieval scenarios. This paper presents a novel approach, termed unpaired robust hashing with noisy labels (URHNL), for zero-shot cross-modal retrieval. Specifically, we developed a zero-shot cross-modal hash retrieval framework that can learn distinct hash codes for different modalities, which is suitable for unpaired cross-modal retrieval scenarios. In addition, it incorporates the sparse constraint on the noise matrix and the low-rank constraint on the recovered label matrix, respectively. These constraints are applied to mitigate the negative effects of noisy labels. Furthermore, we introduce the concept of drag ɛ into the learning process of label semantic embedding, which aims to generate more discriminative hash codes. To improve the similarity of semantic information within hash codes, we consider both intra-modal and inter-modal similarity. A large number of experiments on cross-modal datasets show the effectiveness of the URHNL approach in real and complex zero-shot cross-modal retrieval scenarios. The source code of this word can be found at https://github.com/szq0816/URHNL.

Incomplete Cross-Modal Retrieval with Deep Correlation Transfer

Most cross-modal retrieval methods assume the multi-modal training data is complete and has a one-to-one correspondence. However, in the real world, multi-modal data generally suffers from missing modality information due to the uncertainty of data collection and storage processes, which limits the practical application of existing cross-modal retrieval methods. Although some solutions have been proposed to generate the missing modality data using a single pseudo sample, this may lead to incomplete semantic restoration and sub-optimal retrieval results due to the limited semantic information it provides. To address this challenge, this article proposes an Incomplete Cross-Modal Retrieval with Deep Correlation Transfer (ICMR-DCT) method that can robustly model incomplete multi-modal data and dynamically capture the adjacency semantic correlation for cross-modal retrieval. Specifically, we construct intra-modal graph attention-based auto-encoder to learn modality-invariant representations by performing semantic reconstruction through intra-modality adjacency correlation mining. Then, we design dual cross-modal alignment constraints to project multi-modal representations into a common semantic space, thus bridging the heterogeneous modality gap and enhancing the discriminability of the common representation. We further introduce semantic preservation to enhance adjacency semantic information and achieve cross-modal semantic correlation. Moreover, we propose a nearest-neighbor weighting integration strategy with cross-modal correlation transfer to generate the missing modality data according to inter-modality mapping relations and adjacency correlations between each sample and its neighbors, which improves the robustness of our method against incomplete multi-modal training data. Extensive experiments on three widely tested benchmark datasets demonstrate the superior performance of our method in cross-modal retrieval tasks under both complete and incomplete retrieval scenarios. Our used datasets and source codes are available at https://github.com/shidan0122/DCT.git .

Continual learning for cross-modal image-text retrieval based on domain-selective attention

Cross-modal image-text retrieval (CMITR) has been a high-value research topic for more than a decade. In most of the previous studies, the data for all tasks are trained as a single set. However, in reality, a more likely scenario is that the dataset has multiple tasks and trains them in sequence. The consequence is the limited ability to memorize the old task once a new task arrives; in other words, catastrophic forgetting. To solve this issue, this paper proposes a novel continual learning for cross-modal image-text retrieval (CLCMR) method to alleviate catastrophic forgetting. We construct a multilayer domain-selective attention (MDSA) based network to obtain knowledge from task-relevant and domain-specific attention levels. Moreover, a memory factor has been designed to achieve weight regularization, and a novel memory loss function is utilized to constrain MDSA. The extensive experimental results from multiple datasets (Wikipedia, Pascal Sentence, and PKU XMedianet datasets) demonstrate that CLCMR can effectively alleviate catastrophic forgetting and achieve a superior continual learning ability compared with the state-of-the-art methods.

Cross-Modal Retrieval and Semantic Refinement for Remote Sensing Image Captioning

Two-stage remote sensing image captioning (RSIC) methods have achieved promising results by incorporating additional pre-trained remote sensing tasks to extract supplementary information and improve caption quality. However, these methods face limitations in semantic comprehension, as pre-trained detectors/classifiers are constrained by predefined labels, leading to an oversight of the intricate and diverse details present in remote sensing images (RSIs). Additionally, the handling of auxiliary remote sensing tasks separately can introduce challenges in ensuring seamless integration and alignment with the captioning process. To address these problems, we propose a novel cross-modal retrieval and semantic refinement (CRSR) RSIC method. Specifically, we employ a cross-modal retrieval model to retrieve relevant sentences of each image. The words in these retrieved sentences are then considered as primary semantic information, providing valuable supplementary information for the captioning process. To further enhance the quality of the captions, we introduce a semantic refinement module that refines the primary semantic information, which helps to filter out misleading information and emphasize visually salient semantic information. A Transformer Mapper network is introduced to expand the representation of image features beyond the retrieved supplementary information with learnable queries. Both the refined semantic tokens and visual features are integrated and fed into a cross-modal decoder for caption generation. Through extensive experiments, we demonstrate the superiority of our CRSR method over existing state-of-the-art approaches on the RSICD, the UCM-Captions, and the Sydney-Captions datasets

Cross-Modal Retrieval With Noisy Correspondence via Consistency Refining and Mining.

The success of existing cross-modal retrieval (CMR) methods heavily rely on the assumption that the annotated cross-modal correspondence is faultless. In practice, however, the correspondence of some pairs would be inevitably contaminated during data collection or annotation, thus leading to the so-called Noisy Correspondence (NC) problem. To alleviate the influence of NC, we propose a novel method termed Consistency REfining And Mining (CREAM) by revealing and exploiting the difference between correspondence and consistency. Specifically, the correspondence and the consistency only be coincident for true positive and true negative pairs, while being distinct for false positive and false negative pairs. Based on the observation, CREAM employs a collaborative learning paradigm to detect and rectify the correspondence of positives, and a negative mining approach to explore and utilize the consistency. Thanks to the consistency refining and mining strategy of CREAM, the overfitting on the false positives could be prevented and the consistency rooted in the false negatives could be exploited, thus leading to a robust CMR method. Extensive experiments verify the effectiveness of our method on three image-text benchmarks including Flickr30K, MS-COCO, and Conceptual Captions. Furthermore, we adopt our method into the graph matching task and the results demonstrate the robustness of our method against fine-grained NC problem. The code is available on https://github.com/XLearning-SCU/2024-TIP-CREAM.

Geometric Matching for Cross-Modal Retrieval.

Despite its significant progress, cross-modal retrieval still suffers from one-to-many matching cases, where the multiplicity of semantic instances in another modality could be acquired by a given query. However, existing approaches usually map heterogeneous data into the learned space as deterministic point vectors. In spite of their remarkable performance in matching the most similar instance, such deterministic point embedding suffers from the insufficient representation of rich semantics in one-to-many correspondence. To address the limitations, we intuitively extend a deterministic point into a closed geometry and develop geometric representation learning methods for cross-modal retrieval. Thus, a set of points inside such a geometry could be semantically related to many candidates, and we could effectively capture the semantic uncertainty. We then introduce two types of geometric matching for one-to-many correspondence, i.e., point-to-rectangle matching (dubbed P2RM) and rectangle-to-rectangle matching (termed R2RM). The former treats all retrieved candidates as rectangles with zero volume (equivalent to points) and the query as a box, while the latter encodes all heterogeneous data into rectangles. Therefore, we could evaluate semantic similarity among heterogeneous data by the Euclidean distance from a point to a rectangle or the volume of intersection between two rectangles. Additionally, both strategies could be easily employed for off-the-self approaches and further improve the retrieval performance of baselines. Under various evaluation metrics, extensive experiments and ablation studies on several commonly used datasets, two for image-text matching and two for video-text retrieval, demonstrate our effectiveness and superiority.

Keyword-Based Diverse Image Retrieval With Variational Multiple Instance Graph.

The task of cross-modal image retrieval has recently attracted considerable research attention. In real-world scenarios, keyword-based queries issued by users are usually short and have broad semantics. Therefore, semantic diversity is as important as retrieval accuracy in such user-oriented services, which improves user experience. However, most typical cross-modal image retrieval methods based on single point query embedding inevitably result in low semantic diversity, while existing diverse retrieval approaches frequently lead to low accuracy due to a lack of cross-modal understanding. To address this challenge, we introduce an end-to-end solution termed variational multiple instance graph (VMIG), in which a continuous semantic space is learned to capture diverse query semantics, and the retrieval task is formulated as a multiple instance learning problems to connect diverse features across modalities. Specifically, a query-guided variational autoencoder is employed to model the continuous semantic space instead of learning a single-point embedding. Afterward, multiple instances of the image and query are obtained by sampling in the continuous semantic space and applying multihead attention, respectively. Thereafter, an instance graph is constructed to remove noisy instances and align cross-modal semantics. Finally, heterogeneous modalities are robustly fused under multiple losses. Extensive experiments on two real-world datasets have well verified the effectiveness of our proposed solution in both retrieval accuracy and semantic diversity.

Adversarial pre-optimized graph representation learning with double-order sampling for cross-modal retrieval

Cross-modal retrieval aims at retrieving relevant information across different media data, such as images and text. However, inconsistent distribution of different modal data causes a heterogeneity gap, which becomes an intractable issue. Previous researches aim at finding a semantic share space for various modalities, in which the similarity of them can be calculated by a uniform measurement. Graph representation learning has shown excellent performance in many tasks, like information retrieval. However, it is difficult to construct the most proper cross-modal graph which can better present the semantics of multi-modal data. Besides, the projected representations of various modal data optimized by the supervised class labels and unsupervised graph context learning are insufficiently aligned. To address these issues, this paper presents an Adversarial Pre-optimized Graph Representation Learning (AP-GRL) for cross-modal retrieval. Similar to mainstream methods, AP-GRL aims to learn a shared representation for various modalities. We design an adaptive pre-optimization mechanism to optimize the cross-modal data and their relationships, which is different from previously fixed graph. The raw constructed cross-modal graph is optimized by maximum posterior probability estimation. To fully exploit the graph, we proposed a double-order sampling strategy to take the width and depth into representation learning. Besides, we adopt adversarial learning with Wasserstein distance to reduce domain discrepancy in the domain-invariant representation learning. Extensive experiments on five widely used datasets demonstrate that AP-GRL outperforms state-of-the-art cross-modal retrieval methods.

A Cross-modal image retrieval method based on contrastive learning

A Cross-modal image retrieval method based on contrastive learning

A Cross-Modal Hash Retrieval Method with Fused Triples

Due to the fast retrieval speed and low storage cost, cross-modal hashing has become the primary method for cross-modal retrieval. Since the emergence of deep cross-modal hashing methods, cross-modal retrieval significantly improved. However, the existing cross-modal hash retrieval methods still need to effectively utilize the dataset’s supervisory information and the lack of similarity expression ability. This means that the label information needs to be maximized, and the potential semantic relationship between two modalities cannot be fully explored, thus affecting the judgment of semantic similarity between two modalities. To address these problems, this paper proposes Tri-CMH, a cross-modal hash retrieval method with fused triples, which is an end-to-end modeling framework consisting of two parts: feature extraction and hash learning. Firstly, the multi-modal data are preprocessing into the form of triple groups. The data supervision matrix is constructed so that the samples with labels and their meanings are aggregated together. In contrast, the samples with labels and their opposite meanings are separated, thus avoiding the problem of the under-utilization of supervisory information in the data set and achieving the effect of efficiently utilizing the global supervisory information. Meanwhile, the loss function of the hash learning part is optimized by considering the Hamming distance loss, single-modality internal loss, cross-modality loss, and quantization loss to explicitly constrain semantically similar hash codes and semantically dissimilar hash codes and to improve the model’s ability to judge cross-modality semantic similarity. The method is trained and tested on the IAPR-TC12, MIRFLICKR-25K, and NUS-WIDE datasets, and the experimental evaluation criteria are mAP and PR curve, and the experimental results show the effectiveness and practicality of the method.

Similarity Graph-correlation Reconstruction Network for unsupervised cross-modal hashing

Existing cross-modal hash retrieval methods can simultaneously enhance retrieval speed and reduce storage space. However, these methods face a major challenge in determining the similarity metric between two modalities. Specifically, the accuracy of intra-modal and inter-modal similarity measurements is inadequate, and the large gap between modalities leads to semantic bias. In this paper, we propose a Similarity Graph-correlation Reconstruction Network (SGRN) for unsupervised cross-modal hashing. Particularly, the local relation graph rebasing module is used to filter out graph nodes with weak similarity and associate graph nodes with strong similarity, resulting in fine-grained intra-modal similarity relation graphs. The global relation graph reconstruction module is further strengthens cross-modal correlation and implements fine-grained similarity alignment between modalities. In addition, in order to bridge the modal gap, we combine the similarity representation of real-valued and hash features to design the intra-modal and inter-modal training strategies. SGRN conducted extensive experiments on two cross-modal retrieval datasets, and the experimental results effectively validated the superiority of the proposed method and significantly improved the retrieval performance.

Multimodal Mutual Information Maximization: A Novel Approach for Unsupervised Deep Cross-Modal Hashing.

In this article, we adopt the maximizing mutual information (MI) approach to tackle the problem of unsupervised learning of binary hash codes for efficient cross-modal retrieval. We proposed a novel method, dubbed cross-modal info-max hashing (CMIMH). First, to learn informative representations that can preserve both intramodal and intermodal similarities, we leverage the recent advances in estimating variational lower bound of MI to maximizing the MI between the binary representations and input features and between binary representations of different modalities. By jointly maximizing these MIs under the assumption that the binary representations are modeled by multivariate Bernoulli distributions, we can learn binary representations, which can preserve both intramodal and intermodal similarities, effectively in a mini-batch manner with gradient descent. Furthermore, we find out that trying to minimize the modality gap by learning similar binary representations for the same instance from different modalities could result in less informative representations. Hence, balancing between reducing the modality gap and losing modality-private information is important for the cross-modal retrieval tasks. Quantitative evaluations on standard benchmark datasets demonstrate that the proposed method consistently outperforms other state-of-the-art cross-modal retrieval methods.

Less is Better: Exponential Loss for Cross-Modal Matching

Deep metric learning has become a key component of cross-modal retrieval. By learning to pull the features of matched instances closer while pushing the features of mismatched instances farther away, one can learn highly robust multi-modal representations. Most existing cross-modal retrieval methods leverage vanilla triplet loss to train the network, which cannot adaptively penalize pairs with different hardness. Although various weighting strategies have been designed for unimodal matching tasks, few weighting strategies have been applied to cross-modal tasks due to the specificity of cross-modal tasks. While few weighting strategies are designed for cross-modal scenarios, they usually involve a lot of hyper-parameters, which require a lot of computational resources to fine-tune. In this paper, we introduce a new exponential loss, which can assign appropriate weights to individual positive and negative pairs according to their similarity so that it can adaptively penalize pairs with different hardness. Furthermore, the exponential loss has only two hyper-parameters, making it easier to find the optimal parameters to suit various data distributions in practice. Exponential loss can be universally applied to well-established cross-modal models and further boost their retrieval performance. We exhaustively ablate our method on Image-Text matching, Video-Text matching, as well as unimodal Image matching. Experimental results show that a standard model trained with exponential loss can achieve noticeable performance gains.