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).
Automated radiology report generation aims to expedite the tedious and error-prone reporting process for radiologists. While recent works have made progress, learning to align medical images and textual findings remains challenging due to the relative scarcity of labeled medical data. For example, datasets for this task are much smaller than those used for image captioning in computer vision. In this work, we propose to transfer representations from CLIP, a large-scale pre-trained vision-language model, to better capture cross-modal semantics between images and texts. However, directly applying CLIP is suboptimal due to the domain gap between natural images and radiology. To enable efficient adaptation, we introduce UniCrossAdapter, lightweight adapter modules that are incorporated into CLIP and fine-tuned on the target task while keeping base parameters fixed. The adapters are distributed across modalities and their interaction to enhance vision-language alignment. Experiments on two public datasets demonstrate the effectiveness of our approach, advancing state-of-the-art in radiology report generation. The proposed transfer learning framework provides a means of harnessing semantic knowledge from large-scale pre-trained models to tackle data-scarce medical vision-language tasks. Code is available at https://github.com/chauncey-tow/MRG-CLIP.
In the rapidly evolving field of artificial intelligence, multimodal models, e.g., integrating vision and language into visual-language models (VLMs), have become pivotal for many applications, ranging from image captioning to multimodal search engines. Among these models, the Contrastive Language-Image Pre-training (CLIP) model has demonstrated remarkable performance in understanding and generating nuanced relationships between text and images. However, the conventional training of such models often requires centralized aggregation of vast datasets, posing significant privacy and data governance challenges. To address these concerns, this paper proposes a novel approach that leverages Federated Learning and parameter-efficient adapters, i.e., Low-Rank Adaptation (LoRA), to train VLMs. This methodology preserves data privacy by training models across decentralized data sources and ensures model adaptability and efficiency through LoRA's parameter-efficient fine-tuning. Our approach accelerates training time by up to 34.72 times and requires 2.47 times less memory usage than full fine-tuning.
Multimodal systems have highly complex processing pipelines and are pretrained over large datasets before being fine-tuned for specific tasks such as visual captioning. However, it becomes hard to disentangle what the model learns during the fine-tuning process from what it already knows due to its pretraining. In this work, we learn a probabilistic model using Hybrid Markov Logic Networks (HMLNs) over the training examples by relating symbolic knowledge (extracted from the caption) with visual features (extracted from the image). For a generated caption, we quantify the influence of training examples based on the HMLN distribution using probabilistic inference. We evaluate two types of inference procedures on the MSCOCO dataset for different types of captioning models. Our results show that for BLIP2 (a model that uses a LLM), the fine-tuning may have smaller influence on the knowledge the model has acquired since it may have more general knowledge to perform visual captioning as compared to models that do not use a LLM
Recent progress in (multimodal) large language models ((M)LLMs) has shifted focus from pre-training to inference-time compute scaling and post-training optimization, driven by concerns over limited high-quality real-world data. However, these strategies alone are insufficient for advancing model capabilities. We hypothesize that effective model improvement requires a strong synergy among pre-training, inference-time compute scaling, and post-training optimization. In this paper, we validate this hypothesis in the context of multimodal pre-training for foundation MLLM construction. We introduce Self-Improving cognition (SIcog), a self-learning framework for constructing next-generation foundation MLLMs by imparting multimodal knowledge and enhancing their systematic cognitive capabilities through multimodal pre-training with self-generated data. Specifically, we introduce Chain-of-Description, a step-by-step visual understanding method to improve comprehensive perception, and integrate structured chain-of-thought (CoT) reasoning to support in-depth multimodal reasoning. SIcog first equips a base model with systematic perception and reasoning using minimal external supervision. The enhanced model then generates candidate image captions and CoT-style reasoning responses for unlabeled images and image-question pairs across diverse tasks, which are curated through a self-consistency mechanism. These curated samples are subsequently used for large-scale multimodal pre-training, completing a self-learning cycle that strengthens the model's cognitive foundation. Extensive experiments demonstrate that SIcog produces next-generation foundation MLLMs with substantially improved multimodal cognition, outperforming prevailing pre-training approaches. These findings empirically establish SIcog as a promising framework for realizing a complete self-improving paradigm.
This paper introduces a holistic vision-language foundation model tailored for remote sensing, named Falcon. Falcon offers a unified, prompt-based paradigm that effectively executes comprehensive and complex remote sensing tasks. Falcon demonstrates powerful understanding and reasoning abilities at the image, region, and pixel levels. Specifically, given simple natural language instructions and remote sensing images, Falcon can produce impressive results in text form across 14 distinct tasks, i.e., image classification, object detection, segmentation, image captioning, and etc. To facilitate Falcon's training and empower its representation capacity to encode rich spatial and semantic information, we developed Falcon_SFT, a large-scale, multi-task, instruction-tuning dataset in the field of remote sensing. The Falcon_SFT dataset consists of approximately 78 million high-quality data samples, covering 5.6 million multi-spatial resolution and multi-view remote sensing images with diverse instructions. It features hierarchical annotations and undergoes manual sampling verification to ensure high data quality and reliability. Extensive comparative experiments are conducted, which verify that Falcon achieves remarkable performance over 67 datasets and 14 tasks, despite having only 0.7B parameters. We release the complete dataset, code, and model weights at https://github.com/TianHuiLab/Falcon, hoping to help further develop the open-source community.
Writing Assistants (e.g., Grammarly, Microsoft Copilot) traditionally generate diverse image captions by employing syntactic and semantic variations to describe image components. However, human-written captions prioritize conveying a central message alongside visual descriptions using pragmatic cues. To enhance pragmatic diversity, it is essential to explore alternative ways of communicating these messages in conjunction with visual content. To address this challenge, we propose RONA, a novel prompting strategy for Multi-modal Large Language Models (MLLM) that leverages Coherence Relations as an axis for variation. We demonstrate that RONA generates captions with better overall diversity and ground-truth alignment, compared to MLLM baselines across multiple domains. Our code is available at: https://github.com/aashish2000/RONA
Vision-Language Models (VLMs) have advanced multi-modal tasks like image captioning, visual question answering, and reasoning. However, they often generate hallucinated outputs inconsistent with the visual context or prompt, limiting reliability in critical applications like autonomous driving and medical imaging. Existing studies link hallucination to statistical biases, language priors, and biased feature learning but lack a structured causal understanding. In this work, we introduce a causal perspective to analyze and mitigate hallucination in VLMs. We hypothesize that hallucination arises from unintended direct influences of either the vision or text modality, bypassing proper multi-modal fusion. To address this, we construct a causal graph for VLMs and employ counterfactual analysis to estimate the Natural Direct Effect (NDE) of vision, text, and their cross-modal interaction on the output. We systematically identify and mitigate these unintended direct effects to ensure that responses are primarily driven by genuine multi-modal fusion. Our approach consists of three steps: (1) designing structural causal graphs to distinguish correct fusion pathways from spurious modality shortcuts, (2) estimating modality-specific and cross-modal NDE using perturbed image representations, hallucinated text embeddings, and degraded visual inputs, and (3) implementing a test-time intervention module to dynamically adjust the model's dependence on each modality. Experimental results demonstrate that our method significantly reduces hallucination while preserving task performance, providing a robust and interpretable framework for improving VLM reliability. To enhance accessibility and reproducibility, our code is publicly available at https://github.com/TREE985/Treble-Counterfactual-VLMs.
Existing Video Scene Graph Generation (VidSGG) studies are trained in a fully supervised manner, which requires all frames in a video to be annotated, thereby incurring high annotation cost compared to Image Scene Graph Generation (ImgSGG). Although the annotation cost of VidSGG can be alleviated by adopting a weakly supervised approach commonly used for ImgSGG (WS-ImgSGG) that uses image captions, there are two key reasons that hinder such a naive adoption: 1) Temporality within video captions, i.e., unlike image captions, video captions include temporal markers (e.g., before, while, then, after) that indicate time related details, and 2) Variability in action duration, i.e., unlike human actions in image captions, human actions in video captions unfold over varying duration. To address these issues, we propose a Natural Language-based Video Scene Graph Generation (NL-VSGG) framework that only utilizes the readily available video captions for training a VidSGG model. NL-VSGG consists of two key modules: Temporality-aware Caption Segmentation (TCS) module and Action Duration Variability-aware caption-frame alignment (ADV) module. Specifically, TCS segments the video captions into multiple sentences in a temporal order based on a Large Language Model (LLM), and ADV aligns each segmented sentence with appropriate frames considering the variability in action duration. Our approach leads to a significant enhancement in performance compared to simply applying the WS-ImgSGG pipeline to VidSGG on the Action Genome dataset. As a further benefit of utilizing the video captions as weak supervision, we show that the VidSGG model trained by NL-VSGG is able to predict a broader range of action classes that are not included in the training data, which makes our framework practical in reality.
This paper introduces Multi-Modal Retrieval-Augmented Generation (M^2RAG), a benchmark designed to evaluate the effectiveness of Multi-modal Large Language Models (MLLMs) in leveraging knowledge from multi-modal retrieval documents. The benchmark comprises four tasks: image captioning, multi-modal question answering, multi-modal fact verification, and image reranking. All tasks are set in an open-domain setting, requiring RAG models to retrieve query-relevant information from a multi-modal document collection and use it as input context for RAG modeling. To enhance the context utilization capabilities of MLLMs, we also introduce Multi-Modal Retrieval-Augmented Instruction Tuning (MM-RAIT), an instruction tuning method that optimizes MLLMs within multi-modal contexts. Our experiments show that MM-RAIT improves the performance of RAG systems by enabling them to effectively learn from multi-modal contexts. All data and code are available at https://github.com/NEUIR/M2RAG.
Developing video captioning models is computationally expensive. The dynamic nature of video also complicates the design of multimodal models that can effectively caption these sequences. However, we find that by using minimal computational resources and without complex modifications to address video dynamics, an image-based model can be repurposed to outperform several specialised video captioning systems. Our adapted model demonstrates top tier performance on major benchmarks, ranking 2nd on MSRVTT and MSVD, and 3rd on VATEX. We transform it into a competitive video captioner by post training a typical image captioning model BLIP2 with only 6,000 video text pairs and simply concatenating frames (significantly fewer data than other methods), which use 2.5 to 144 million pairs. From a resource optimization perspective, this video captioning study focuses on three fundamental factors: optimizing model scale, maximizing data efficiency, and incorporating reinforcement learning. This extensive study demonstrates that a lightweight, image based adaptation strategy can rival state-of-the-art video captioning systems, offering a practical solution for low-resource scenarios.
The evaluation of image captions, looking at both linguistic fluency and semantic correspondence to visual contents, has witnessed a significant effort. Still, despite advancements such as the CLIPScore metric, multilingual captioning evaluation has remained relatively unexplored. This work presents several strategies, and extensive experiments, related to evaluating CLIPScore variants in multilingual settings. To address the lack of multilingual test data, we consider two different strategies: (1) using quality aware machine-translated datasets with human judgements, and (2) re-purposing multilingual datasets that target semantic inference and reasoning. Our results highlight the potential of finetuned multilingual models to generalize across languages and to handle complex linguistic challenges. Tests with machine-translated data show that multilingual CLIPScore models can maintain a high correlation with human judgements across different languages, and additional tests with natively multilingual and multicultural data further attest to the high-quality assessments.
The continuous operation of Earth-orbiting satellites generates vast and ever-growing archives of Remote Sensing (RS) images. Natural language presents an intuitive interface for accessing, querying, and interpreting the data from such archives. However, existing Vision-Language Models (VLMs) are predominantly trained on web-scraped, noisy image-text data, exhibiting limited exposure to the specialized domain of RS. This deficiency results in poor performance on RS-specific tasks, as commonly used datasets often lack detailed, scientifically accurate textual descriptions and instead emphasize solely on attributes like date and location. To bridge this critical gap, we introduce GAIA, a novel dataset designed for multi-scale, multi-sensor, and multi-modal RS image analysis. GAIA comprises of 205,150 meticulously curated RS image-text pairs, representing a diverse range of RS modalities associated to different spatial resolutions. Unlike existing vision-language datasets in RS, GAIA specifically focuses on capturing a diverse range of RS applications, providing unique information about environmental changes, natural disasters, and various other dynamic phenomena. The dataset provides a spatially and temporally balanced distribution, spanning across the globe, covering the last 25 years with a balanced temporal distribution of observations. GAIA's construction involved a two-stage process: (1) targeted web-scraping of images and accompanying text from reputable RS-related sources, and (2) generation of five high-quality, scientifically grounded synthetic captions for each image using carefully crafted prompts that leverage the advanced vision-language capabilities of GPT-4o. Our extensive experiments, including fine-tuning of CLIP and BLIP2 models, demonstrate that GAIA significantly improves performance on RS image classification, cross-modal retrieval and image captioning tasks.
Multimodal foundation models (MFMs) have demonstrated significant success in tasks such as visual captioning, question answering, and image-text retrieval. However, these models face inherent limitations due to their finite internal capacity, which restricts their ability to process extended temporal sequences, a crucial requirement for comprehensive video and audio analysis. To overcome these challenges, we introduce a specialized cognitive module, temporal working memory (TWM), which aims to enhance the temporal modeling capabilities of MFMs. It selectively retains task-relevant information across temporal dimensions, ensuring that critical details are preserved throughout the processing of video and audio content. The TWM uses a query-guided attention approach to focus on the most informative multimodal segments within temporal sequences. By retaining only the most relevant content, TWM optimizes the use of the model's limited capacity, enhancing its temporal modeling ability. This plug-and-play module can be easily integrated into existing MFMs. With our TWM, nine state-of-the-art models exhibit significant performance improvements across tasks such as video captioning, question answering, and video-text retrieval. By enhancing temporal modeling, TWM extends the capability of MFMs to handle complex, time-sensitive data effectively. Our code is available at https://github.com/xid32/NAACL_2025_TWM.
Efforts to connect LiDAR data with text, such as LidarCLIP, have primarily focused on embedding 3D point clouds into CLIP text-image space. However, these approaches rely on 3D point clouds, which present challenges in encoding efficiency and neural network processing. With the advent of advanced LiDAR sensors like Ouster OS1, which, in addition to 3D point clouds, produce fixed resolution depth, signal, and ambient panoramic 2D images, new opportunities emerge for LiDAR based tasks. In this work, we propose an alternative approach to connect LiDAR data with text by leveraging 2D imagery generated by the OS1 sensor instead of 3D point clouds. Using the Florence 2 large model in a zero-shot setting, we perform image captioning and object detection. Our experiments demonstrate that Florence 2 generates more informative captions and achieves superior performance in object detection tasks compared to existing methods like CLIP. By combining advanced LiDAR sensor data with a large pre-trained model, our approach provides a robust and accurate solution for challenging detection scenarios, including real-time applications requiring high accuracy and robustness.
Multi-modal Large Language Models (MLLMs) excel in vision-language tasks but remain vulnerable to visual adversarial perturbations that can induce hallucinations, manipulate responses, or bypass safety mechanisms. Existing methods seek to mitigate these risks by applying constrained adversarial fine-tuning to CLIP vision encoders on ImageNet-scale data, ensuring their generalization ability is preserved. However, this limited adversarial training restricts robustness and broader generalization. In this work, we explore an alternative approach of leveraging existing vision classification models that have been adversarially pre-trained on large-scale data. Our analysis reveals two principal contributions: (1) the extensive scale and diversity of adversarial pre-training enables these models to demonstrate superior robustness against diverse adversarial threats, ranging from imperceptible perturbations to advanced jailbreaking attempts, without requiring additional adversarial training, and (2) end-to-end MLLM integration with these robust models facilitates enhanced adaptation of language components to robust visual features, outperforming existing plug-and-play methodologies on complex reasoning tasks. Through systematic evaluation across visual question-answering, image captioning, and jail-break attacks, we demonstrate that MLLMs trained with these robust models achieve superior adversarial robustness while maintaining favorable clean performance. Our framework achieves 2x and 1.5x average robustness gains in captioning and VQA tasks, respectively, and delivers over 10% improvement against jailbreak attacks. Code and pretrained models will be available at https://github.com/HashmatShadab/Robust-LLaVA.
Memes have emerged as a powerful form of communication, integrating visual and textual elements to convey humor, satire, and cultural messages. Existing research has focused primarily on aspects such as emotion classification, meme generation, propagation, interpretation, figurative language, and sociolinguistics, but has often overlooked deeper meme comprehension and meme-text retrieval. To address these gaps, this study introduces ClassicMemes-50-templates (CM50), a large-scale dataset consisting of over 33,000 memes, centered around 50 popular meme templates. We also present an automated knowledge-grounded annotation pipeline leveraging large vision-language models to produce high-quality image captions, meme captions, and literary device labels overcoming the labor intensive demands of manual annotation. Additionally, we propose a meme-text retrieval CLIP model (mtrCLIP) that utilizes cross-modal embedding to enhance meme analysis, significantly improving retrieval performance. Our contributions include:(1) a novel dataset for large-scale meme study, (2) a scalable meme annotation framework, and (3) a fine-tuned CLIP for meme-text retrieval, all aimed at advancing the understanding and analysis of memes at scale.
Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities in understanding and describing visual content, achieving state-of-the-art performance across various vision-language tasks. However, these models often generate descriptions containing objects or details that are absent in the input image, a phenomenon commonly known as hallucination. Our work investigates the key reasons behind this issue by analyzing the pattern of self-attention in transformer layers. We find that hallucinations often arise from the progressive weakening of attention weight to visual tokens in the deeper layers of the LLM. Some previous works naively boost the attention of all visual tokens to mitigate this issue, resulting in suboptimal hallucination reduction. To address this, we identify two critical sets of visual tokens that facilitate the transfer of visual information from the vision encoder to the LLM. Local tokens encode grounded information about objects present in an image, while summary tokens capture the overall aggregated representation of the image. Importantly, these two sets of tokens require different levels of weight enhancement. To this end, we propose \textbf{PAINT} (\textbf{P}aying \textbf{A}ttention to \textbf{IN}formed \textbf{T}okens), a plug-and-play framework that intervenes in the self-attention mechanism of the LLM, selectively boosting the attention weights of local and summary tokens with experimentally learned margins. Evaluation on the MSCOCO image captioning dataset demonstrate that our approach reduces hallucination rates by up to 62.3\% compared to baseline models while maintaining accuracy. Code is available at \href{https://github.com/hasanar1f/PAINT}{https://github.com/hasanar1f/PAINT}
Purpose: Surgical workflow analysis is crucial for improving surgical efficiency and safety. However, previous studies rely heavily on large-scale annotated datasets, posing challenges in cost, scalability, and reliance on expert annotations. To address this, we propose Surg-FTDA (Few-shot Text-driven Adaptation), designed to handle various surgical workflow analysis tasks with minimal paired image-label data. Methods: Our approach has two key components. First, Few-shot selection-based modality alignment selects a small subset of images and aligns their embeddings with text embeddings from the downstream task, bridging the modality gap. Second, Text-driven adaptation leverages only text data to train a decoder, eliminating the need for paired image-text data. This decoder is then applied to aligned image embeddings, enabling image-related tasks without explicit image-text pairs. Results: We evaluate our approach to generative tasks (image captioning) and discriminative tasks (triplet recognition and phase recognition). Results show that Surg-FTDA outperforms baselines and generalizes well across downstream tasks. Conclusion: We propose a text-driven adaptation approach that mitigates the modality gap and handles multiple downstream tasks in surgical workflow analysis, with minimal reliance on large annotated datasets. The code and dataset will be released in https://github.com/CAMMA-public/Surg-FTDA
Automated audio captioning is a task that generates textual descriptions for audio content, and recent studies have explored using visual information to enhance captioning quality. However, current methods often fail to effectively fuse audio and visual data, missing important semantic cues from each modality. To address this, we introduce LAVCap, a large language model (LLM)-based audio-visual captioning framework that effectively integrates visual information with audio to improve audio captioning performance. LAVCap employs an optimal transport-based alignment loss to bridge the modality gap between audio and visual features, enabling more effective semantic extraction. Additionally, we propose an optimal transport attention module that enhances audio-visual fusion using an optimal transport assignment map. Combined with the optimal training strategy, experimental results demonstrate that each component of our framework is effective. LAVCap outperforms existing state-of-the-art methods on the AudioCaps dataset, without relying on large datasets or post-processing. Code is available at https://github.com/NAVER-INTEL-Co-Lab/gaudi-lavcap.