Multimodal Large Language Models (MLLMs) have showcased impressive skills in tasks related to visual understanding and reasoning. Yet, their widespread application faces obstacles due to the high computational demands during both the training and inference phases, restricting their use to a limited audience within the research and user communities. In this paper, we investigate the design aspects of Multimodal Small Language Models (MSLMs) and propose an efficient multimodal assistant named Mipha, which is designed to create synergy among various aspects: visual representation, language models, and optimization strategies. We show that without increasing the volume of training data, our Mipha-3B outperforms the state-of-the-art large MLLMs, especially LLaVA-1.5-13B, on multiple benchmarks. Through detailed discussion, we provide insights and guidelines for developing strong MSLMs that rival the capabilities of MLLMs. Our code is available at https://github.com/zhuyiche/llava-phi.
In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.
We present DeepSeek-VL, an open-source Vision-Language (VL) Model designed for real-world vision and language understanding applications. Our approach is structured around three key dimensions: We strive to ensure our data is diverse, scalable, and extensively covers real-world scenarios including web screenshots, PDFs, OCR, charts, and knowledge-based content, aiming for a comprehensive representation of practical contexts. Further, we create a use case taxonomy from real user scenarios and construct an instruction tuning dataset accordingly. The fine-tuning with this dataset substantially improves the model's user experience in practical applications. Considering efficiency and the demands of most real-world scenarios, DeepSeek-VL incorporates a hybrid vision encoder that efficiently processes high-resolution images (1024 x 1024), while maintaining a relatively low computational overhead. This design choice ensures the model's ability to capture critical semantic and detailed information across various visual tasks. We posit that a proficient Vision-Language Model should, foremost, possess strong language abilities. To ensure the preservation of LLM capabilities during pretraining, we investigate an effective VL pretraining strategy by integrating LLM training from the beginning and carefully managing the competitive dynamics observed between vision and language modalities. The DeepSeek-VL family (both 1.3B and 7B models) showcases superior user experiences as a vision-language chatbot in real-world applications, achieving state-of-the-art or competitive performance across a wide range of visual-language benchmarks at the same model size while maintaining robust performance on language-centric benchmarks. We have made both 1.3B and 7B models publicly accessible to foster innovations based on this foundation model.
Vision-extended LLMs have made significant strides in Visual Question Answering (VQA). Despite these advancements, VLLMs still encounter substantial difficulties in handling queries involving long-tail entities, with a tendency to produce erroneous or hallucinated responses. In this work, we introduce a novel evaluative benchmark named \textbf{SnapNTell}, specifically tailored for entity-centric VQA. This task aims to test the models' capabilities in identifying entities and providing detailed, entity-specific knowledge. We have developed the \textbf{SnapNTell Dataset}, distinct from traditional VQA datasets: (1) It encompasses a wide range of categorized entities, each represented by images and explicitly named in the answers; (2) It features QA pairs that require extensive knowledge for accurate responses. The dataset is organized into 22 major categories, containing 7,568 unique entities in total. For each entity, we curated 10 illustrative images and crafted 10 knowledge-intensive QA pairs. To address this novel task, we devised a scalable, efficient, and transparent retrieval-augmented multimodal LLM. Our approach markedly outperforms existing methods on the SnapNTell dataset, achieving a 66.5\% improvement in the BELURT score. We will soon make the dataset and the source code publicly accessible.
This paper focuses on the challenge of answering questions in scenarios that are composed of rich and complex dynamic audio-visual components. Although existing Multimodal Large Language Models (MLLMs) can respond to audio-visual content, these responses are sometimes ambiguous and fail to describe specific audio-visual events. To overcome this limitation, we introduce the CAT, which enhances MLLM in three ways: 1) besides straightforwardly bridging audio and video, we design a clue aggregator that aggregates question-related clues in dynamic audio-visual scenarios to enrich the detailed knowledge required for large language models. 2) CAT is trained on a mixed multimodal dataset, allowing direct application in audio-visual scenarios. Notably, we collect an audio-visual joint instruction dataset named AVinstruct, to further enhance the capacity of CAT to model cross-semantic correlations. 3) we propose AI-assisted ambiguity-aware direct preference optimization, a strategy specialized in retraining the model to favor the non-ambiguity response and improve the ability to localize specific audio-visual objects. Extensive experimental results demonstrate that CAT outperforms existing methods on multimodal tasks, especially in Audio-Visual Question Answering (AVQA) tasks. The codes and the collected instructions are released at https://github.com/rikeilong/Bay-CAT.
This paper introduces the novel task of multimodal puzzle solving, framed within the context of visual question-answering. We present a new dataset, AlgoPuzzleVQA designed to challenge and evaluate the capabilities of multimodal language models in solving algorithmic puzzles that necessitate both visual understanding, language understanding, and complex algorithmic reasoning. We create the puzzles to encompass a diverse array of mathematical and algorithmic topics such as boolean logic, combinatorics, graph theory, optimization, search, etc., aiming to evaluate the gap between visual data interpretation and algorithmic problem-solving skills. The dataset is generated automatically from code authored by humans. All our puzzles have exact solutions that can be found from the algorithm without tedious human calculations. It ensures that our dataset can be scaled up arbitrarily in terms of reasoning complexity and dataset size. Our investigation reveals that large language models (LLMs) such as GPT4V and Gemini exhibit limited performance in puzzle-solving tasks. We find that their performance is near random in a multi-choice question-answering setup for a significant number of puzzles. The findings emphasize the challenges of integrating visual, language, and algorithmic knowledge for solving complex reasoning problems.
Multimodal Large Language Models (MLLMs) have experienced significant advancements recently. Nevertheless, challenges persist in the accurate recognition and comprehension of intricate details within high-resolution images. Despite being indispensable for the development of robust MLLMs, this area remains underinvestigated. To tackle this challenge, our work introduces InfiMM-HD, a novel architecture specifically designed for processing images of different resolutions with low computational overhead. This innovation facilitates the enlargement of MLLMs to higher-resolution capabilities. InfiMM-HD incorporates a cross-attention module and visual windows to reduce computation costs. By integrating this architectural design with a four-stage training pipeline, our model attains improved visual perception efficiently and cost-effectively. Empirical study underscores the robustness and effectiveness of InfiMM-HD, opening new avenues for exploration in related areas. Codes and models can be found at https://huggingface.co/Infi-MM/infimm-hd
Leveraging pre-trained visual language models has become a widely adopted approach for improving performance in downstream visual question answering (VQA) applications. However, in the specialized field of medical VQA, the scarcity of available data poses a significant barrier to achieving reliable model generalization. Numerous methods have been proposed to enhance model generalization, addressing the issue from data-centric and model-centric perspectives. Data augmentation techniques are commonly employed to enrich the dataset, while various regularization approaches aim to prevent model overfitting, especially when training on limited data samples. In this paper, we introduce a method that incorporates gradient-guided parameter perturbations to the visual encoder of the multimodality model during both pre-training and fine-tuning phases, to improve model generalization for downstream medical VQA tasks. The small perturbation is adaptively generated by aligning with the direction of the moving average gradient in the optimization landscape, which is opposite to the directions of the optimizer's historical updates. It is subsequently injected into the model's visual encoder. The results show that, even with a significantly smaller pre-training image caption dataset, our approach achieves competitive outcomes on both VQA-RAD and SLAKE datasets.
From content moderation to wildlife conservation, the number of applications that require models to recognize nuanced or subjective visual concepts is growing. Traditionally, developing classifiers for such concepts requires substantial manual effort measured in hours, days, or even months to identify and annotate data needed for training. Even with recently proposed Agile Modeling techniques, which enable rapid bootstrapping of image classifiers, users are still required to spend 30 minutes or more of monotonous, repetitive data labeling just to train a single classifier. Drawing on Fiske's Cognitive Miser theory, we propose a new framework that alleviates manual effort by replacing human labeling with natural language interactions, reducing the total effort required to define a concept by an order of magnitude: from labeling 2,000 images to only 100 plus some natural language interactions. Our framework leverages recent advances in foundation models, both large language models and vision-language models, to carve out the concept space through conversation and by automatically labeling training data points. Most importantly, our framework eliminates the need for crowd-sourced annotations. Moreover, our framework ultimately produces lightweight classification models that are deployable in cost-sensitive scenarios. Across 15 subjective concepts and across 2 public image classification datasets, our trained models outperform traditional Agile Modeling as well as state-of-the-art zero-shot classification models like ALIGN, CLIP, CuPL, and large visual question-answering models like PaLI-X.
Medical vision-language models (VLMs) combine computer vision (CV) and natural language processing (NLP) to analyze visual and textual medical data. Our paper reviews recent advancements in developing VLMs specialized for healthcare, focusing on models designed for medical report generation and visual question answering (VQA). We provide background on NLP and CV, explaining how techniques from both fields are integrated into VLMs to enable learning from multimodal data. Key areas we address include the exploration of medical vision-language datasets, in-depth analyses of architectures and pre-training strategies employed in recent noteworthy medical VLMs, and comprehensive discussion on evaluation metrics for assessing VLMs' performance in medical report generation and VQA. We also highlight current challenges and propose future directions, including enhancing clinical validity and addressing patient privacy concerns. Overall, our review summarizes recent progress in developing VLMs to harness multimodal medical data for improved healthcare applications.
Despite remarkable progress, existing multimodal large language models (MLLMs) are still inferior in granular visual recognition. Contrary to previous works, we study this problem from the perspective of image resolution, and reveal that a combination of low- and high-resolution visual features can effectively mitigate this shortcoming. Based on this observation, we propose a novel and efficient method for MLLMs, termed Mixture-of-Resolution Adaptation (MRA). In particular, MRA adopts two visual pathways for images with different resolutions, where high-resolution visual information is embedded into the low-resolution pathway via the novel mixture-of-resolution adapters (MR-Adapters). This design also greatly reduces the input sequence length of MLLMs. To validate MRA, we apply it to a recent MLLM called LLaVA, and term the new model LLaVA-HR. We conduct extensive experiments on 11 vision-language (VL) tasks, which show that LLaVA-HR outperforms existing MLLMs on 8 VL tasks, e.g., +9.4% on TextVQA. More importantly, both training and inference of LLaVA-HR remain efficient with MRA, e.g., 20 training hours and 3$\times$ inference speed than LLaVA-1.5. Source codes are released at: https://github.com/luogen1996/LLaVA-HR.
The integration of learning and reasoning is high on the research agenda in AI. Nevertheless, there is only a little attention to use existing background knowledge for reasoning about partially observed scenes to answer questions about the scene. Yet, we as humans use such knowledge frequently to infer plausible answers to visual questions (by eliminating all inconsistent ones). Such knowledge often comes in the form of constraints about objects and it tends to be highly domain or environment-specific. We contribute a novel benchmark called CLEVR-POC for reasoning-intensive visual question answering (VQA) in partially observable environments under constraints. In CLEVR-POC, knowledge in the form of logical constraints needs to be leveraged to generate plausible answers to questions about a hidden object in a given partial scene. For instance, if one has the knowledge that all cups are colored either red, green or blue and that there is only one green cup, it becomes possible to deduce the color of an occluded cup as either red or blue, provided that all other cups, including the green one, are observed. Through experiments, we observe that the low performance of pre-trained vision language models like CLIP (~ 22%) and a large language model (LLM) like GPT-4 (~ 46%) on CLEVR-POC ascertains the necessity for frameworks that can handle reasoning-intensive tasks where environment-specific background knowledge is available and crucial. Furthermore, our demonstration illustrates that a neuro-symbolic model, which integrates an LLM like GPT-4 with a visual perception network and a formal logical reasoner, exhibits exceptional performance on CLEVR-POC.
Open-world generalization requires robotic systems to have a profound understanding of the physical world and the user command to solve diverse and complex tasks. While the recent advancement in vision-language models (VLMs) has offered unprecedented opportunities to solve open-world problems, how to leverage their capabilities to control robots remains a grand challenge. In this paper, we introduce Marking Open-world Keypoint Affordances (MOKA), an approach that employs VLMs to solve robotic manipulation tasks specified by free-form language instructions. Central to our approach is a compact point-based representation of affordance, which bridges the VLM's predictions on observed images and the robot's actions in the physical world. By prompting the pre-trained VLM, our approach utilizes the VLM's commonsense knowledge and concept understanding acquired from broad data sources to predict affordances and generate motions. To facilitate the VLM's reasoning in zero-shot and few-shot manners, we propose a visual prompting technique that annotates marks on images, converting affordance reasoning into a series of visual question-answering problems that are solvable by the VLM. We further explore methods to enhance performance with robot experiences collected by MOKA through in-context learning and policy distillation. We evaluate and analyze MOKA's performance on various table-top manipulation tasks including tool use, deformable body manipulation, and object rearrangement.
Understanding the contents of multimodal documents is essential to accurately extract relevant evidence and use it for reasoning. Existing document understanding models tend to generate answers with a single word or phrase directly, ignoring the source document's evidence and lacking interpretability. In this work, we address the lack of step-wise capabilities through data augmentation and extension. Specifically, We use Multi-modal Large Language Models (MLLMs), which have strong visual understanding and reasoning abilities, as data generators to generate step-wise question-and-answer pairs for document images and use a high-performance LLM as the error detector to filter out noisy data. This step-wise data generation pipeline is implemented using both template-based and few-shot methods. We then use the generated high-quality data to train a humanized document understanding and reasoning model, specifically designed to solve complex questions that require reasoning or multi-hop question answering, dubbed DocAssistant. Experimental results demonstrate the effectiveness and application value of step-wise generation, showing a 5 improvement on InfoVQA with complex layouts and a 7 improvement on ChartQA with complex reasoning, compared to directly generated answers. We hope our work highlights the potential of synthetic data and encourages further exploration of multi-modal document reasoning capabilities.
We present the All-Seeing Project V2: a new model and dataset designed for understanding object relations in images. Specifically, we propose the All-Seeing Model V2 (ASMv2) that integrates the formulation of text generation, object localization, and relation comprehension into a relation conversation (ReC) task. Leveraging this unified task, our model excels not only in perceiving and recognizing all objects within the image but also in grasping the intricate relation graph between them, diminishing the relation hallucination often encountered by Multi-modal Large Language Models (MLLMs). To facilitate training and evaluation of MLLMs in relation understanding, we created the first high-quality ReC dataset ({AS-V2) which is aligned with the format of standard instruction tuning data. In addition, we design a new benchmark, termed Circular-based Relation Probing Evaluation (CRPE) for comprehensively evaluating the relation comprehension capabilities of MLLMs. Notably, our ASMv2 achieves an overall accuracy of 52.04 on this relation-aware benchmark, surpassing the 43.14 of LLaVA-1.5 by a large margin. We hope that our work can inspire more future research and contribute to the evolution towards artificial general intelligence. Our project is released at https://github.com/OpenGVLab/all-seeing.
Large Vision-Language Models (LVLMs), despite their recent success, are hardly comprehensively tested for their cognitive abilities. Inspired by the prevalent use of the "Cookie Theft" task in human cognition test, we propose a novel evaluation benchmark to evaluate high-level cognitive ability of LVLMs using images with rich semantics. It defines eight reasoning capabilities and consists of an image description task and a visual question answering task. Our evaluation on well-known LVLMs shows that there is still a large gap in cognitive ability between LVLMs and humans.
Visual question answering (VQA) is crucial for promoting surgical education. In practice, the needs of trainees are constantly evolving, such as learning more surgical types, adapting to different robots, and learning new surgical instruments and techniques for various surgeries. However, patient data privacy often restricts the availability of old data when updating the model, necessitating an exemplar-free continual learning (CL) setup. Prior CL studies overlooked two vital problems in the surgical domain: 1) large domain shifts from diverse surgical operations collected from multiple sources, and 2) severe data imbalance arising from the uneven presence of surgical instruments or activities. This paper proposes addressing these problems with a multimodal large language model (LLM) and an adaptive weight assignment methodology. We first develop a new multi-teacher CL framework that leverages a multimodal LLM as the additional teacher. The strong generalization ability of the LLM can bridge the knowledge gap when domain shifts and data imbalances occur. We then put forth a novel data processing method that transforms complex LLM embeddings into logits compatible with our CL framework. We further design an adaptive weight assignment approach that balances the generalization ability of the LLM and the domain expertise of the old CL model. Finally, to comprehensively test the effectiveness of our proposed method, we have also constructed two new surgical VQA datasets that are largely different from existing ones and could be valuable resources for future research. Extensive experimental results on the tested datasets demonstrate the superiority of our method to other advanced CL schemes.
"A data scientist is tasked with developing a low-cost surgical VQA system for a 2-month workshop. Due to data sensitivity, she collects 50 hours of surgical video from a hospital, requiring two months for privacy approvals. Privacy restrictions prevent uploading data to platforms like ChatGPT, so she assembles one annotator and a medical expert to manually create QA pairs. This process takes three weeks and costs over $10,000. The trained model provides accurate responses within the limited data scope but lacks broader generalizability, completing the project in 3 months." To simplify the challenges presented in the scenario above. In this paper, we replace the image input with text for Vision-language training. Inspired by prior noise injection methods to reduce modality gaps, we introduce Adaptive ranged cosine Similarity injected noise (ArcSin). First, we introduce an innovative adaptive noise scale that effectively generates the textual elements with more variability while preserving the original text feature's integrity. Second, a similarity pool strategy is employed, expanding the domain generalization potential by broadening the overall noise scale. This dual strategy effectively broadens the scope of the original domain while safeguarding content integrity. Our empirical results demonstrate that these models closely rival those trained on images in terms of performance. Specifically, our method exhibits substantial improvements over the previous state-of-the-art, achieving gains of 1.9 and 1.1 CIDEr points in S-Cap and M-Cap, respectively. Additionally, we observe increases of 0.5 percentage points (pp), 1.4 pp, and 1.4 pp in accuracy for VQA, VQA-E, and VE, respectively, pushing the boundaries of what is achievable within the constraints of image-trained model benchmarks.
Visual commonsense contains knowledge about object properties, relationships, and behaviors in visual data. Discovering visual commonsense can provide a more comprehensive and richer understanding of images, and enhance the reasoning and decision-making capabilities of computer vision systems. However, the visual commonsense defined in existing visual commonsense discovery studies is coarse-grained and incomplete. In this work, we draw inspiration from a commonsense knowledge base ConceptNet in natural language processing, and systematically define the types of visual commonsense. Based on this, we introduce a new task, Visual Commonsense Discovery (VCD), aiming to extract fine-grained commonsense of different types contained within different objects in the image. We accordingly construct a dataset (VCDD) from Visual Genome and ConceptNet for VCD, featuring over 100,000 images and 14 million object-commonsense pairs. We furthermore propose a generative model (VCDM) that integrates a vision-language model with instruction tuning to tackle VCD. Automatic and human evaluations demonstrate VCDM's proficiency in VCD, particularly outperforming GPT-4V in implicit commonsense discovery. The value of VCD is further demonstrated by its application to two downstream tasks, including visual commonsense evaluation and visual question answering. The data and code will be made available on GitHub.
In 3D Visual Question Answering (3D VQA), the scarcity of fully annotated data and limited visual content diversity hampers the generalization to novel scenes and 3D concepts (e.g., only around 800 scenes are utilized in ScanQA and SQA dataset). Current approaches resort supplement 3D reasoning with 2D information. However, these methods face challenges: either they use top-down 2D views that introduce overly complex and sometimes question-irrelevant visual clues, or they rely on globally aggregated scene/image-level representations from 2D VLMs, losing the fine-grained vision-language correlations. To overcome these limitations, our approach utilizes question-conditional 2D view selection procedure, pinpointing semantically relevant 2D inputs for crucial visual clues. We then integrate this 2D knowledge into the 3D-VQA system via a two-branch Transformer structure. This structure, featuring a Twin-Transformer design, compactly combines 2D and 3D modalities and captures fine-grained correlations between modalities, allowing them mutually augmenting each other. Integrating proposed mechanisms above, we present BridgeQA, that offers a fresh perspective on multi-modal transformer-based architectures for 3D-VQA. Experiments validate that BridgeQA achieves state-of-the-art on 3D-VQA datasets and significantly outperforms existing solutions. Code is available at $\href{https://github.com/matthewdm0816/BridgeQA}{\text{this URL}}$.