Retrieval-Augmented Generation (RAG) has emerged as a powerful paradigm for enhancing the capabilities of large language models. However, existing RAG evaluation predominantly focuses on text retrieval and relies on opaque, end-to-end assessments of generated outputs. To address these limitations, we introduce mmRAG, a modular benchmark designed for evaluating multi-modal RAG systems. Our benchmark integrates queries from six diverse question-answering datasets spanning text, tables, and knowledge graphs, which we uniformly convert into retrievable documents. To enable direct, granular evaluation of individual RAG components -- such as the accuracy of retrieval and query routing -- beyond end-to-end generation quality, we follow standard information retrieval procedures to annotate document relevance and derive dataset relevance. We establish baseline performance by evaluating a wide range of RAG implementations on mmRAG.
China has a long and rich history, encompassing a vast cultural heritage that includes diverse multimodal information, such as silk patterns, Dunhuang murals, and their associated historical narratives. Cross-modal retrieval plays a pivotal role in understanding and interpreting Chinese cultural heritage by bridging visual and textual modalities to enable accurate text-to-image and image-to-text retrieval. However, despite the growing interest in multimodal research, there is a lack of specialized datasets dedicated to Chinese cultural heritage, limiting the development and evaluation of cross-modal learning models in this domain. To address this gap, we propose a multimodal dataset named CulTi, which contains 5,726 image-text pairs extracted from two series of professional documents, respectively related to ancient Chinese silk and Dunhuang murals. Compared to existing general-domain multimodal datasets, CulTi presents a challenge for cross-modal retrieval: the difficulty of local alignment between intricate decorative motifs and specialized textual descriptions. To address this challenge, we propose LACLIP, a training-free local alignment strategy built upon a fine-tuned Chinese-CLIP. LACLIP enhances the alignment of global textual descriptions with local visual regions by computing weighted similarity scores during inference. Experimental results on CulTi demonstrate that LACLIP significantly outperforms existing models in cross-modal retrieval, particularly in handling fine-grained semantic associations within Chinese cultural heritage.
Visual Document Retrieval (VDR) is an emerging research area that focuses on encoding and retrieving document images directly, bypassing the dependence on Optical Character Recognition (OCR) for document search. A recent advance in VDR was introduced by ColPali, which significantly improved retrieval effectiveness through a late interaction mechanism. ColPali's approach demonstrated substantial performance gains over existing baselines that do not use late interaction on an established benchmark. In this study, we investigate the reproducibility and replicability of VDR methods with and without late interaction mechanisms by systematically evaluating their performance across multiple pre-trained vision-language models. Our findings confirm that late interaction yields considerable improvements in retrieval effectiveness; however, it also introduces computational inefficiencies during inference. Additionally, we examine the adaptability of VDR models to textual inputs and assess their robustness across text-intensive datasets within the proposed benchmark, particularly when scaling the indexing mechanism. Furthermore, our research investigates the specific contributions of late interaction by looking into query-patch matching in the context of visual document retrieval. We find that although query tokens cannot explicitly match image patches as in the text retrieval scenario, they tend to match the patch contains visually similar tokens or their surrounding patches.
Leaf disease identification plays a pivotal role in smart agriculture. However, many existing studies still struggle to integrate image and textual modalities to compensate for each other's limitations. Furthermore, many of these approaches rely on pretraining with constrained datasets such as ImageNet, which lack domain-specific information. We propose SCOLD (Soft-target COntrastive learning for Leaf Disease identification), a context-aware vision-language foundation model tailored to address these challenges for agricultural tasks. SCOLD is developed using a diverse corpus of plant leaf images and corresponding symptom descriptions, comprising over 186,000 image-caption pairs aligned with 97 unique concepts. Through task-agnostic pretraining, SCOLD leverages contextual soft targets to mitigate overconfidence in contrastive learning by smoothing labels, thereby improving model generalization and robustness on fine-grained classification tasks. Experimental results demonstrate that SCOLD outperforms existing vision-language models such as OpenAI-CLIP-L, BioCLIP, and SigLIP2 across several benchmarks, including zero-shot and few-shot classification, image-text retrieval, and image classification, while maintaining a competitive parameter footprint. Ablation studies further highlight SCOLD's effectiveness in contrast to its counterparts. The proposed approach significantly advances the agricultural vision-language foundation model, offering strong performance with minimal or no supervised fine-tuning. This work lays a solid groundwork for future research on models trained with long-form and simplified contexts, tasks involving class ambiguity, and multi-modal systems for intelligent plant disease diagnostics. The code for this study is available at https://huggingface.co/enalis/scold
The Query-By-Document (QBD) problem is an information retrieval problem where the query is a document, and the retrieved candidates are documents that match the query document, often in a domain or query specific manner. This can be crucial for tasks such as patent matching, legal or compliance case retrieval, and academic literature review. Existing retrieval methods, including keyword search and document embeddings, can be optimized with domain-specific datasets to improve QBD search performance. However, creating these domain-specific datasets is often costly and time-consuming. Our work introduces a process to generate custom QBD-search datasets and compares a set of methods to use in this problem, which we refer to as QBD-RankedDatagen. We provide a comparative analysis of our proposed methods in terms of cost, speed, and the human interface with the domain experts. The methods we compare leverage Large Language Models (LLMs) which can incorporate domain expert input to produce document scores and rankings, as well as explanations for human review. The process and methods for it that we present can significantly reduce human effort in dataset creation for custom domains while still obtaining sufficient expert knowledge for tuning retrieval models. We evaluate our methods on QBD datasets from the Text Retrieval Conference (TREC) and finetune the parameters of the BM25 model -- which is used in many industrial-strength search engines like OpenSearch -- using the generated data.
Contrastive Language-Image Pre-training (CLIP) excels in multimodal tasks such as image-text retrieval and zero-shot classification but struggles with fine-grained understanding due to its focus on coarse-grained short captions. To address this, we propose Fine-Grained CLIP (FG-CLIP), which enhances fine-grained understanding through three key innovations. First, we leverage large multimodal models to generate 1.6 billion long caption-image pairs for capturing global-level semantic details. Second, a high-quality dataset is constructed with 12 million images and 40 million region-specific bounding boxes aligned with detailed captions to ensure precise, context-rich representations. Third, 10 million hard fine-grained negative samples are incorporated to improve the model's ability to distinguish subtle semantic differences. We construct a comprehensive dataset, termed FineHARD, by integrating high-quality region-specific annotations with hard fine-grained negative samples. Corresponding training methods are meticulously designed for these data. Extensive experiments demonstrate that FG-CLIP outperforms the original CLIP and other state-of-the-art methods across various downstream tasks, including fine-grained understanding, open-vocabulary object detection, image-text retrieval, and general multimodal benchmarks. These results highlight FG-CLIP's effectiveness in capturing fine-grained image details and improving overall model performance. The data, code, and models are available at https://github.com/360CVGroup/FG-CLIP.
The Contrastive Language-Image Pre-training (CLIP) framework has become a widely used approach for multimodal representation learning, particularly in image-text retrieval and clustering. However, its efficacy is constrained by three key limitations: (1) text token truncation, (2) isolated image-text encoding, and (3) deficient compositionality due to bag-of-words behavior. While recent Multimodal Large Language Models (MLLMs) have demonstrated significant advances in generalized vision-language understanding, their potential for learning transferable multimodal representations remains underexplored.In this work, we present UniME (Universal Multimodal Embedding), a novel two-stage framework that leverages MLLMs to learn discriminative representations for diverse downstream tasks. In the first stage, we perform textual discriminative knowledge distillation from a powerful LLM-based teacher model to enhance the embedding capability of the MLLM\'s language component. In the second stage, we introduce hard negative enhanced instruction tuning to further advance discriminative representation learning. Specifically, we initially mitigate false negative contamination and then sample multiple hard negatives per instance within each batch, forcing the model to focus on challenging samples. This approach not only improves discriminative power but also enhances instruction-following ability in downstream tasks. We conduct extensive experiments on the MMEB benchmark and multiple retrieval tasks, including short and long caption retrieval and compositional retrieval. Results demonstrate that UniME achieves consistent performance improvement across all tasks, exhibiting superior discriminative and compositional capabilities.
We introduce CameraBench, a large-scale dataset and benchmark designed to assess and improve camera motion understanding. CameraBench consists of ~3,000 diverse internet videos, annotated by experts through a rigorous multi-stage quality control process. One of our contributions is a taxonomy of camera motion primitives, designed in collaboration with cinematographers. We find, for example, that some motions like "follow" (or tracking) require understanding scene content like moving subjects. We conduct a large-scale human study to quantify human annotation performance, revealing that domain expertise and tutorial-based training can significantly enhance accuracy. For example, a novice may confuse zoom-in (a change of intrinsics) with translating forward (a change of extrinsics), but can be trained to differentiate the two. Using CameraBench, we evaluate Structure-from-Motion (SfM) and Video-Language Models (VLMs), finding that SfM models struggle to capture semantic primitives that depend on scene content, while VLMs struggle to capture geometric primitives that require precise estimation of trajectories. We then fine-tune a generative VLM on CameraBench to achieve the best of both worlds and showcase its applications, including motion-augmented captioning, video question answering, and video-text retrieval. We hope our taxonomy, benchmark, and tutorials will drive future efforts towards the ultimate goal of understanding camera motions in any video.
Recent advancement in large-scale Artificial Intelligence (AI) models offering multimodal services have become foundational in AI systems, making them prime targets for model theft. Existing methods select Out-of-Distribution (OoD) data as backdoor watermarks and retrain the original model for copyright protection. However, existing methods are susceptible to malicious detection and forgery by adversaries, resulting in watermark evasion. In this work, we propose Model-\underline{ag}nostic Black-box Backdoor W\underline{ate}rmarking Framework (AGATE) to address stealthiness and robustness challenges in multimodal model copyright protection. Specifically, we propose an adversarial trigger generation method to generate stealthy adversarial triggers from ordinary dataset, providing visual fidelity while inducing semantic shifts. To alleviate the issue of anomaly detection among model outputs, we propose a post-transform module to correct the model output by narrowing the distance between adversarial trigger image embedding and text embedding. Subsequently, a two-phase watermark verification is proposed to judge whether the current model infringes by comparing the two results with and without the transform module. Consequently, we consistently outperform state-of-the-art methods across five datasets in the downstream tasks of multimodal image-text retrieval and image classification. Additionally, we validated the robustness of AGATE under two adversarial attack scenarios.
Table retrieval is essential for accessing information stored in structured tabular formats; however, it remains less explored than text retrieval. The content of the table primarily consists of phrases and words, which include a large number of entities, such as time, locations, persons, and organizations. Entities are well-studied in the context of text retrieval, but there is a noticeable lack of research on their applications in table retrieval. In this work, we explore how to leverage entities in tables to improve retrieval performance. First, we investigate the important role of entities in table retrieval from a statistical perspective and propose an entity-enhanced training framework. Subsequently, we use the type of entities to highlight entities instead of introducing an external knowledge base. Moreover, we design an interaction paradigm based on entity representations. Our proposed framework is plug-and-play and flexible, making it easy to integrate into existing table retriever training processes. Empirical results on two table retrieval benchmarks, NQ-TABLES and OTT-QA, show that our proposed framework is both simple and effective in enhancing existing retrievers. We also conduct extensive analyses to confirm the efficacy of different components. Overall, our work provides a promising direction for elevating table retrieval, enlightening future research in this area.
The automatic generation of radiology reports has emerged as a promising solution to reduce a time-consuming task and accurately capture critical disease-relevant findings in X-ray images. Previous approaches for radiology report generation have shown impressive performance. However, there remains significant potential to improve accuracy by ensuring that retrieved reports contain disease-relevant findings similar to those in the X-ray images and by refining generated reports. In this study, we propose a Disease-aware image-text Alignment and self-correcting Re-alignment for Trustworthy radiology report generation (DART) framework. In the first stage, we generate initial reports based on image-to-text retrieval with disease-matching, embedding both images and texts in a shared embedding space through contrastive learning. This approach ensures the retrieval of reports with similar disease-relevant findings that closely align with the input X-ray images. In the second stage, we further enhance the initial reports by introducing a self-correction module that re-aligns them with the X-ray images. Our proposed framework achieves state-of-the-art results on two widely used benchmarks, surpassing previous approaches in both report generation and clinical efficacy metrics, thereby enhancing the trustworthiness of radiology reports.
Cross-modal retrieval (CMR) is a fundamental task in multimedia research, focused on retrieving semantically relevant targets across different modalities. While traditional CMR methods match text and image via embedding-based similarity calculations, recent advancements in pre-trained generative models have established generative retrieval as a promising alternative. This paradigm assigns each target a unique identifier and leverages a generative model to directly predict identifiers corresponding to input queries without explicit indexing. Despite its great potential, current generative CMR approaches still face semantic information insufficiency in both identifier construction and generation processes. To address these limitations, we propose a novel unified Semantic-enhanced generative Cross-mOdal REtrieval framework (SemCORE), designed to unleash the semantic understanding capabilities in generative cross-modal retrieval task. Specifically, we first construct a Structured natural language IDentifier (SID) that effectively aligns target identifiers with generative models optimized for natural language comprehension and generation. Furthermore, we introduce a Generative Semantic Verification (GSV) strategy enabling fine-grained target discrimination. Additionally, to the best of our knowledge, SemCORE is the first framework to simultaneously consider both text-to-image and image-to-text retrieval tasks within generative cross-modal retrieval. Extensive experiments demonstrate that our framework outperforms state-of-the-art generative cross-modal retrieval methods. Notably, SemCORE achieves substantial improvements across benchmark datasets, with an average increase of 8.65 points in Recall@1 for text-to-image retrieval.
Visual understanding is inherently contextual -- what we focus on in an image depends on the task at hand. For instance, given an image of a person holding a bouquet of flowers, we may focus on either the person such as their clothing, or the type of flowers, depending on the context of interest. Yet, most existing image encoding paradigms represent an image as a fixed, generic feature vector, overlooking the potential needs of prioritizing varying visual information for different downstream use cases. In this work, we introduce FocalLens, a conditional visual encoding method that produces different representations for the same image based on the context of interest, expressed flexibly through natural language. We leverage vision instruction tuning data and contrastively finetune a pretrained vision encoder to take natural language instructions as additional inputs for producing conditional image representations. Extensive experiments validate that conditional image representation from FocalLens better pronounce the visual features of interest compared to generic features produced by standard vision encoders like CLIP. In addition, we show FocalLens further leads to performance improvements on a range of downstream tasks including image-image retrieval, image classification, and image-text retrieval, with an average gain of 5 and 10 points on the challenging SugarCrepe and MMVP-VLM benchmarks, respectively.
In this work, we introduce long-video masked-embedding autoencoders (LV-MAE), a self-supervised learning framework for long video representation. Our approach treats short- and long-span dependencies as two separate tasks. Such decoupling allows for a more intuitive video processing where short-span spatiotemporal primitives are first encoded and are then used to capture long-range dependencies across consecutive video segments. To achieve this, we leverage advanced off-the-shelf multimodal encoders to extract representations from short segments within the long video, followed by pre-training a masked-embedding autoencoder capturing high-level interactions across segments. LV-MAE is highly efficient to train and enables the processing of much longer videos by alleviating the constraint on the number of input frames. Furthermore, unlike existing methods that typically pre-train on short-video datasets, our approach offers self-supervised pre-training using long video samples (e.g., 20+ minutes video clips) at scale. Using LV-MAE representations, we achieve state-of-the-art results on three long-video benchmarks -- LVU, COIN, and Breakfast -- employing only a simple classification head for either attentive or linear probing. Finally, to assess LV-MAE pre-training and visualize its reconstruction quality, we leverage the video-language aligned space of short video representations to monitor LV-MAE through video-text retrieval.
Video-text retrieval has been stuck in the information mismatch caused by personalized and inadequate textual descriptions of videos. The substantial information gap between the two modalities hinders an effective cross-modal representation alignment, resulting in ambiguous retrieval results. Although text rewriting methods have been proposed to broaden text expressions, the modality gap remains significant, as the text representation space is hardly expanded with insufficient semantic enrichment.Instead, this paper turns to enhancing visual presentation, bridging video expression closer to textual representation via caption generation and thereby facilitating video-text matching.While multimodal large language models (mLLM) have shown a powerful capability to convert video content into text, carefully crafted prompts are essential to ensure the reasonableness and completeness of the generated captions. Therefore, this paper proposes an automatic caption enhancement method that improves expression quality and mitigates empiricism in augmented captions through self-learning.Additionally, an expertized caption selection mechanism is designed and introduced to customize augmented captions for each video, further exploring the utilization potential of caption augmentation.Our method is entirely data-driven, which not only dispenses with heavy data collection and computation workload but also improves self-adaptability by circumventing lexicon dependence and introducing personalized matching. The superiority of our method is validated by state-of-the-art results on various benchmarks, specifically achieving Top-1 recall accuracy of 68.5% on MSR-VTT, 68.1% on MSVD, and 62.0% on DiDeMo. Our code is publicly available at https://github.com/CaryXiang/ECA4VTR.
Video-text retrieval, the task of retrieving videos based on a textual query or vice versa, is of paramount importance for video understanding and multimodal information retrieval. Recent methods in this area rely primarily on visual and textual features and often ignore audio, although it helps enhance overall comprehension of video content. Moreover, traditional models that incorporate audio blindly utilize the audio input regardless of whether it is useful or not, resulting in suboptimal video representation. To address these limitations, we propose a novel video-text retrieval framework, Audio-guided VIdeo representation learning with GATEd attention (AVIGATE), that effectively leverages audio cues through a gated attention mechanism that selectively filters out uninformative audio signals. In addition, we propose an adaptive margin-based contrastive loss to deal with the inherently unclear positive-negative relationship between video and text, which facilitates learning better video-text alignment. Our extensive experiments demonstrate that AVIGATE achieves state-of-the-art performance on all the public benchmarks.
Contrastive language-audio pre-training (CLAP) has addressed audio-language tasks such as audio-text retrieval by aligning audio and text in a common feature space. While CLAP addresses general audio-language tasks, its audio features do not generalize well in audio tasks. In contrast, self-supervised learning (SSL) models learn general-purpose audio features that perform well in diverse audio tasks. We pursue representation learning that can be widely used in audio applications and hypothesize that a method that learns both general audio features and CLAP features should achieve our goal, which we call a general-purpose audio-language representation. To implement our hypothesis, we propose M2D2, a second-generation masked modeling duo (M2D) that combines an SSL M2D and CLAP. M2D2 learns two types of features using two modalities (audio and text) in a two-stage training process. It also utilizes advanced LLM-based sentence embeddings in CLAP training for powerful semantic supervision. In the first stage, M2D2 learns generalizable audio features from M2D and CLAP, where CLAP aligns the features with the fine LLM-based semantic embeddings. In the second stage, it learns CLAP features using the audio features learned from the LLM-based embeddings. Through these pre-training stages, M2D2 should enhance generalizability and performance in its audio and CLAP features. Experiments validated that M2D2 achieves effective general-purpose audio-language representation, highlighted with SOTA fine-tuning mAP of 49.0 for AudioSet, SOTA performance in music tasks, and top-level performance in audio-language tasks.
Table retrieval, essential for accessing information through tabular data, is less explored compared to text retrieval. The row/column structure and distinct fields of tables (including titles, headers, and cells) present unique challenges. For example, different table fields have varying matching preferences: cells may favor finer-grained (word/phrase level) matching over broader (sentence/passage level) matching due to their fragmented and detailed nature, unlike titles. This necessitates a table-specific retriever to accommodate the various matching needs of each table field. Therefore, we introduce a Table-tailored HYbrid Matching rEtriever (THYME), which approaches table retrieval from a field-aware hybrid matching perspective. Empirical results on two table retrieval benchmarks, NQ-TABLES and OTT-QA, show that THYME significantly outperforms state-of-the-art baselines. Comprehensive analyses confirm the differing matching preferences across table fields and validate the design of THYME.
Despite that deep learning (DL) methods have presented tremendous potential in many medical image analysis tasks, the practical applications of medical DL models are limited due to the lack of enough data samples with manual annotations. By noting that the clinical radiology examinations are associated with radiology reports that describe the images, we propose to develop a foundation model for multi-model head MRI by using contrastive learning on the images and the corresponding radiology findings. In particular, a contrastive learning framework is proposed, where a mixed syntax and semantic similarity matching metric is integrated to reduce the thirst of extreme large dataset in conventional contrastive learning framework. Our proposed similarity enhanced contrastive language image pretraining (SeLIP) is able to effectively extract more useful features. Experiments revealed that our proposed SeLIP performs well in many downstream tasks including image-text retrieval task, classification task, and image segmentation, which highlights the importance of considering the similarities among texts describing different images in developing medical image foundation models.
Vision-Language Models (VLMs) have recently emerged as powerful tools, excelling in tasks that integrate visual and textual comprehension, such as image captioning, visual question answering, and image-text retrieval. However, existing benchmarks for VLMs include spatial components, which often fail to isolate spatial reasoning from related tasks such as object detection or semantic comprehension. In this paper, we address these deficiencies with a multi-faceted approach towards understanding spatial reasoning. Informed by the diverse and multi-dimensional nature of human spatial reasoning abilities, we present a detailed analysis that first delineates the core elements of spatial reasoning: spatial relations, orientation and navigation, mental rotation, and spatial visualization, and then assesses the performance of these models in both synthetic and real-world images, bridging controlled and naturalistic contexts. We analyze 13 state-of-the-art Vision-Language Models, uncovering pivotal insights into their spatial reasoning performance. Our results reveal profound shortcomings in current VLMs, with average accuracy across the 13 models approximating random chance, highlighting spatial reasoning as a persistent obstacle. This work not only exposes the pressing need to advance spatial reasoning within VLMs but also establishes a solid platform for future exploration. Code available on GitHub (https://github.com/stogiannidis/srbench) and dataset available on HuggingFace (https://huggingface.co/datasets/stogiannidis/srbench).