GuessWhich is an engaging visual dialogue game that involves interaction between a Questioner Bot (QBot) and an Answer Bot (ABot) in the context of image-guessing. In this game, QBot's objective is to locate a concealed image solely through a series of visually related questions posed to ABot. However, effectively modeling visually related reasoning in QBot's decision-making process poses a significant challenge. Current approaches either lack visual information or rely on a single real image sampled at each round as decoding context, both of which are inadequate for visual reasoning. To address this limitation, we propose a novel approach that focuses on visually related reasoning through the use of a mental model of the undisclosed image. Within this framework, QBot learns to represent mental imagery, enabling robust visual reasoning by tracking the dialogue state. The dialogue state comprises a collection of representations of mental imagery, as well as representations of the entities involved in the conversation. At each round, QBot engages in visually related reasoning using the dialogue state to construct an internal representation, generate relevant questions, and update both the dialogue state and internal representation upon receiving an answer. Our experimental results on the VisDial datasets (v0.5, 0.9, and 1.0) demonstrate the effectiveness of our proposed model, as it achieves new state-of-the-art performance across all metrics and datasets, surpassing previous state-of-the-art models. Codes and datasets from our experiments are freely available at \href{https://github.com/xubuvd/GuessWhich}.
Significant research efforts have been made to scale and improve vision-language model (VLM) training approaches. Yet, with an ever-growing number of benchmarks, researchers are tasked with the heavy burden of implementing each protocol, bearing a non-trivial computational cost, and making sense of how all these benchmarks translate into meaningful axes of progress. To facilitate a systematic evaluation of VLM progress, we introduce UniBench: a unified implementation of 50+ VLM benchmarks spanning a comprehensive range of carefully categorized capabilities from object recognition to spatial awareness, counting, and much more. We showcase the utility of UniBench for measuring progress by evaluating nearly 60 publicly available vision-language models, trained on scales of up to 12.8B samples. We find that while scaling training data or model size can boost many vision-language model capabilities, scaling offers little benefit for reasoning or relations. Surprisingly, we also discover today's best VLMs struggle on simple digit recognition and counting tasks, e.g. MNIST, which much simpler networks can solve. Where scale falls short, we find that more precise interventions, such as data quality or tailored-learning objectives offer more promise. For practitioners, we also offer guidance on selecting a suitable VLM for a given application. Finally, we release an easy-to-run UniBench code-base with the full set of 50+ benchmarks and comparisons across 59 models as well as a distilled, representative set of benchmarks that runs in 5 minutes on a single GPU.
Recognizing and disentangling visual attributes from objects is a foundation to many computer vision applications. While large vision language representations like CLIP had largely resolved the task of zero-shot object recognition, zero-shot visual attribute recognition remains a challenge because CLIP's contrastively-learned vision-language representation cannot effectively capture object-attribute dependencies. In this paper, we target this weakness and propose a sentence generation-based retrieval formulation for attribute recognition that is novel in 1) explicitly modeling a to-be-measured and retrieved object-attribute relation as a conditional probability graph, which converts the recognition problem into a dependency-sensitive language-modeling problem, and 2) applying a large pretrained Vision-Language Model (VLM) on this reformulation and naturally distilling its knowledge of image-object-attribute relations to use towards attribute recognition. Specifically, for each attribute to be recognized on an image, we measure the visual-conditioned probability of generating a short sentence encoding the attribute's relation to objects on the image. Unlike contrastive retrieval, which measures likelihood by globally aligning elements of the sentence to the image, generative retrieval is sensitive to the order and dependency of objects and attributes in the sentence. We demonstrate through experiments that generative retrieval consistently outperforms contrastive retrieval on two visual reasoning datasets, Visual Attribute in the Wild (VAW), and our newly-proposed Visual Genome Attribute Ranking (VGARank).
Large language models (LLMs) have transformed the development of embodied intelligence. By providing a few contextual demonstrations, developers can utilize the extensive internal knowledge of LLMs to effortlessly translate complex tasks described in abstract language into sequences of code snippets, which will serve as the execution logic for embodied agents. However, this paper uncovers a significant backdoor security threat within this process and introduces a novel method called \method{}. By poisoning just a few contextual demonstrations, attackers can covertly compromise the contextual environment of a black-box LLM, prompting it to generate programs with context-dependent defects. These programs appear logically sound but contain defects that can activate and induce unintended behaviors when the operational agent encounters specific triggers in its interactive environment. To compromise the LLM's contextual environment, we employ adversarial in-context generation to optimize poisoned demonstrations, where an LLM judge evaluates these poisoned prompts, reporting to an additional LLM that iteratively optimizes the demonstration in a two-player adversarial game using chain-of-thought reasoning. To enable context-dependent behaviors in downstream agents, we implement a dual-modality activation strategy that controls both the generation and execution of program defects through textual and visual triggers. We expand the scope of our attack by developing five program defect modes that compromise key aspects of confidentiality, integrity, and availability in embodied agents. To validate the effectiveness of our approach, we conducted extensive experiments across various tasks, including robot planning, robot manipulation, and compositional visual reasoning. Additionally, we demonstrate the potential impact of our approach by successfully attacking real-world autonomous driving systems.
Compositional visual reasoning methods, which translate a complex query into a structured composition of feasible visual tasks, have exhibited a strong potential in complicated multi-modal tasks. Empowered by recent advances in large language models (LLMs), this multi-modal challenge has been brought to a new stage by treating LLMs as few-shot/zero-shot planners, i.e., vision-language (VL) programming. Such methods, despite their numerous merits, suffer from challenges due to LLM planning mistakes or inaccuracy of visual execution modules, lagging behind the non-compositional models. In this work, we devise a "plug-and-play" method, ExoViP, to correct errors in both the planning and execution stages through introspective verification. We employ verification modules as "exoskeletons" to enhance current VL programming schemes. Specifically, our proposed verification module utilizes a mixture of three sub-verifiers to validate predictions after each reasoning step, subsequently calibrating the visual module predictions and refining the reasoning trace planned by LLMs. Experimental results on two representative VL programming methods showcase consistent improvements on five compositional reasoning tasks on standard benchmarks. In light of this, we believe that ExoViP can foster better performance and generalization on open-domain multi-modal challenges.
Human-object interactions (HOI) detection aims at capturing human-object pairs in images and corresponding actions. It is an important step toward high-level visual reasoning and scene understanding. However, due to the natural bias from the real world, existing methods mostly struggle with rare human-object pairs and lead to sub-optimal results. Recently, with the development of the generative model, a straightforward approach is to construct a more balanced dataset based on a group of supplementary samples. Unfortunately, there is a significant domain gap between the generated data and the original data, and simply merging the generated images into the original dataset cannot significantly boost the performance. To alleviate the above problem, we present a novel model-agnostic framework called \textbf{C}ontext-\textbf{E}nhanced \textbf{F}eature \textbf{A}lignment (CEFA) module, which can effectively align the generated data with the original data at the feature level and bridge the domain gap. Specifically, CEFA consists of a feature alignment module and a context enhancement module. On one hand, considering the crucial role of human-object pairs information in HOI tasks, the feature alignment module aligns the human-object pairs by aggregating instance information. On the other hand, to mitigate the issue of losing important context information caused by the traditional discriminator-style alignment method, we employ a context-enhanced image reconstruction module to improve the model's learning ability of contextual cues. Extensive experiments have shown that our method can serve as a plug-and-play module to improve the detection performance of HOI models on rare categories\footnote{https://github.com/LijunZhang01/CEFA}.
Automatic furniture layout is long desired for convenient interior design. Leveraging the remarkable visual reasoning capabilities of multimodal large language models (MLLMs), recent methods address layout generation in a static manner, lacking the feedback-driven refinement essential for interactive user engagement. We introduce Chat2Layout, a novel interactive furniture layout generation system that extends the functionality of MLLMs into the realm of interactive layout design. To achieve this, we establish a unified vision-question paradigm for in-context learning, enabling seamless communication with MLLMs to steer their behavior without altering model weights. Within this framework, we present a novel training-free visual prompting mechanism. This involves a visual-text prompting technique that assist MLLMs in reasoning about plausible layout plans, followed by an Offline-to-Online search (O2O-Search) method, which automatically identifies the minimal set of informative references to provide exemplars for visual-text prompting. By employing an agent system with MLLMs as the core controller, we enable bidirectional interaction. The agent not only comprehends the 3D environment and user requirements through linguistic and visual perception but also plans tasks and reasons about actions to generate and arrange furniture within the virtual space. Furthermore, the agent iteratively updates based on visual feedback from execution results. Experimental results demonstrate that our approach facilitates language-interactive generation and arrangement for diverse and complex 3D furniture.
Visual question answering (VQA) is the task of providing accurate answers to natural language questions based on visual input. Programmatic VQA (PVQA) models have been gaining attention recently. These use large language models (LLMs) to formulate executable programs that address questions requiring complex visual reasoning. However, there are challenges in enabling LLMs to comprehend the usage of image processing modules and generate relevant code. To overcome these challenges, this paper introduces PyramidCoder, a novel prompting framework for PVQA models. PyramidCoder consists of three hierarchical levels, each serving a distinct purpose: query rephrasing, code generation, and answer aggregation. Notably, PyramidCoder utilizes a single frozen LLM and pre-defined prompts at each level, eliminating the need for additional training and ensuring flexibility across various LLM architectures. Compared to the state-of-the-art PVQA model, our approach improves accuracy by at least 0.5% on the GQA dataset, 1.4% on the VQAv2 dataset, and 2.9% on the NLVR2 dataset.
Visual reasoning, as a prominent research area, plays a crucial role in AI by facilitating concept formation and interaction with the world. However, current works are usually carried out separately on small datasets thus lacking generalization ability. Through rigorous evaluation of diverse benchmarks, we demonstrate the shortcomings of existing ad-hoc methods in achieving cross-domain reasoning and their tendency to data bias fitting. In this paper, we revisit visual reasoning with a two-stage perspective: (1) symbolization and (2) logical reasoning given symbols or their representations. We find that the reasoning stage is better at generalization than symbolization. Thus, it is more efficient to implement symbolization via separated encoders for different data domains while using a shared reasoner. Given our findings, we establish design principles for visual reasoning frameworks following the separated symbolization and shared reasoning. The proposed two-stage framework achieves impressive generalization ability on various visual reasoning tasks, including puzzles, physical prediction, and visual question answering (VQA), encompassing both 2D and 3D modalities. We believe our insights will pave the way for generalizable visual reasoning.
We introduce Wonderful Team, a multi-agent Vision Large Language Model (VLLM) framework for executing high-level robotic planning in a zero-shot regime. In our context, zero-shot high-level planning means that for a novel environment, we provide a VLLM with an image of the robot's surroundings and a task description, and the VLLM outputs the sequence of actions necessary for the robot to complete the task. Unlike previous methods for high-level visual planning for robotic manipulation, our method uses VLLMs for the entire planning process, enabling a more tightly integrated loop between perception, control, and planning. As a result, Wonderful Team's performance on real-world semantic and physical planning tasks often exceeds methods that rely on separate vision systems. For example, we see an average 40% success rate improvement on VimaBench over prior methods such as NLaP, an average 30% improvement over Trajectory Generators on tasks from the Trajectory Generator paper, including drawing and wiping a plate, and an average 70% improvement over Trajectory Generators on a new set of semantic reasoning tasks including environment rearrangement with implicit linguistic constraints. We hope these results highlight the rapid improvements of VLLMs in the past year, and motivate the community to consider VLLMs as an option for some high-level robotic planning problems in the future.
The nature of abstract reasoning is a matter of debate. Modern artificial neural network (ANN) models, like large language models, demonstrate impressive success when tested on abstract reasoning problems. However, it has been argued that their success reflects some form of memorization of similar problems (data contamination) rather than a general-purpose abstract reasoning capability. This concern is supported by evidence of brittleness, and the requirement of extensive training. In our study, we explored whether abstract reasoning can be achieved using the toolbox of ANNs, without prior training. Specifically, we studied an ANN model in which the weights of a naive network are optimized during the solution of the problem, using the problem data itself, rather than any prior knowledge. We tested this modeling approach on visual reasoning problems and found that it performs relatively well. Crucially, this success does not rely on memorization of similar problems. We further suggest an explanation of how it works. Finally, as problem solving is performed by changing the ANN weights, we explored the connection between problem solving and the accumulation of knowledge in the ANNs.
This paper investigates visual analogical reasoning in large multimodal models (LMMs) compared to human adults and children. A "visual analogy" is an abstract rule inferred from one image and applied to another. While benchmarks exist for testing visual reasoning in LMMs, they require advanced skills and omit basic visual analogies that even young children can make. Inspired by developmental psychology, we propose a new benchmark of 1,400 visual transformations of everyday objects to test LMMs on visual analogical reasoning and compare them to children and adults. We structure the evaluation into three stages: identifying what changed (e.g., color, number, etc.), how it changed (e.g., added one object), and applying the rule to new scenarios. Our findings show that while models like GPT-4V, LLaVA-1.5, and MANTIS identify the "what" effectively, they struggle with quantifying the "how" and extrapolating this rule to new objects. In contrast, children and adults exhibit much stronger analogical reasoning at all three stages. Additionally, the strongest tested model, GPT-4V, performs better in tasks involving simple visual attributes like color and size, correlating with quicker human adult response times. Conversely, more complex tasks such as number, rotation, and reflection, which necessitate extensive cognitive processing and understanding of the 3D physical world, present more significant challenges. Altogether, these findings highlight the limitations of training models on data that primarily consists of 2D images and text.
The rapid development of Large Language Models (LLMs) has brought impressive advancements across various tasks. However, despite these achievements, LLMs still pose inherent safety risks, especially in the context of jailbreak attacks. Most existing jailbreak methods follow an input-level manipulation paradigm to bypass safety mechanisms. Yet, as alignment techniques improve, such attacks are becoming increasingly detectable. In this work, we identify an underexplored threat vector: the model's internal reasoning process, which can be manipulated to elicit harmful outputs in a more stealthy way. To explore this overlooked attack surface, we propose a novel black-box jailbreak attack method, Analyzing-based Jailbreak (ABJ). ABJ comprises two independent attack paths: textual and visual reasoning attacks, which exploit the model's multimodal reasoning capabilities to bypass safety mechanisms, comprehensively exposing vulnerabilities in its reasoning chain. We conduct extensive experiments on ABJ across various open-source and closed-source LLMs, VLMs, and RLMs. In particular, ABJ achieves high attack success rate (ASR) (82.1% on GPT-4o-2024-11-20) with exceptional attack efficiency (AE) among all target models, showcasing its remarkable attack effectiveness, transferability, and efficiency. Our work reveals a new type of safety risk and highlights the urgent need to mitigate implicit vulnerabilities in the model's reasoning process.
Recent advancements have introduced multiple vision-language models (VLMs) demonstrating impressive commonsense reasoning across various domains. Despite their individual capabilities, the potential of synergizing these complementary VLMs remains underexplored. The Cola Framework addresses this by showcasing how a large language model (LLM) can efficiently coordinate multiple VLMs through natural language communication, leveraging their distinct strengths. We have verified this claim on the challenging A-OKVQA dataset, confirming the effectiveness of such coordination. Building on this, our study investigates whether the same methodology can be applied to surveillance videos for action recognition. Specifically, we explore if leveraging the combined knowledge base of VLMs and LLM can effectively deduce actions from a video when presented with only a few selectively important frames and minimal temporal information. Our experiments demonstrate that LLM, when coordinating different VLMs, can successfully recognize patterns and deduce actions in various scenarios despite the weak temporal signals. However, our findings suggest that to enhance this approach as a viable alternative solution, integrating a stronger temporal signal and exposing the models to slightly more frames would be beneficial.
Visual Language Models (VLMs) are essential for various tasks, particularly visual reasoning tasks, due to their robust multi-modal information integration, visual reasoning capabilities, and contextual awareness. However, existing \VLMs{}' visual spatial reasoning capabilities are often inadequate, struggling even with basic tasks such as distinguishing left from right. To address this, we propose the \ours{} model, designed to enhance the visual spatial reasoning abilities of VLMS. ZeroVLM employs Zero-1-to-3, a 3D reconstruction model for obtaining different views of the input images and incorporates a prompting mechanism to further improve visual spatial reasoning. Experimental results on four visual spatial reasoning datasets show that our \ours{} achieves up to 19.48% accuracy improvement, which indicates the effectiveness of the 3D reconstruction and prompting mechanisms of our ZeroVLM.
Recent advancements in Multimodal Large Language Models (MLLMs) have revolutionized the field of vision-language understanding by integrating visual perception capabilities into Large Language Models (LLMs). The prevailing trend in this field involves the utilization of a vision encoder derived from vision-language contrastive learning (CL), showing expertise in capturing overall representations while facing difficulties in capturing detailed local patterns. In this work, we focus on enhancing the visual representations for MLLMs by combining high-frequency and detailed visual representations, obtained through masked image modeling (MIM), with semantically-enriched low-frequency representations captured by CL. To achieve this goal, we introduce X-Former which is a lightweight transformer module designed to exploit the complementary strengths of CL and MIM through an innovative interaction mechanism. Specifically, X-Former first bootstraps vision-language representation learning and multimodal-to-multimodal generative learning from two frozen vision encoders, i.e., CLIP-ViT (CL-based) and MAE-ViT (MIM-based). It further bootstraps vision-to-language generative learning from a frozen LLM to ensure visual features from X-Former can be interpreted by the LLM. To demonstrate the effectiveness of our approach, we assess its performance on tasks demanding detailed visual understanding. Extensive evaluations indicate that X-Former excels in visual reasoning tasks involving both structural and semantic categories in the GQA dataset. Assessment on fine-grained visual perception benchmark further confirms its superior capabilities in visual understanding.
We consider the problem of finding spatial configurations of multiple objects in images, e.g., a mobile inspection robot is tasked to localize abandoned tools on the floor. We define the spatial configuration of objects by first-order logic in terms of relations and attributes. A neuro-symbolic program matches the logic formulas to probabilistic object proposals for the given image, provided by language-vision models by querying them for the symbols. This work is the first to combine neuro-symbolic programming (reasoning) and language-vision models (learning) to find spatial configurations of objects in images in an open world setting. We show the effectiveness by finding abandoned tools on floors and leaking pipes. We find that most prediction errors are due to biases in the language-vision model.
Large language models (LLMs) and multimodal models (MMs) have exhibited impressive capabilities in various domains, particularly in general language understanding and visual reasoning. However, these models, trained on massive data, may not be finely optimized for specific tasks triggered by instructions. Continual instruction tuning is crucial to adapt a large model to evolving tasks and domains, ensuring their effectiveness and relevance across a wide range of applications. In the context of continual instruction tuning, where models are sequentially trained on different tasks, catastrophic forgetting can occur, leading to performance degradation on previously learned tasks. This work addresses the catastrophic forgetting in continual instruction learning through a switching mechanism for routing computations to parameter-efficient tuned models. We demonstrate the effectiveness of our method through experiments on continual instruction tuning of different natural language generation tasks and vision-language tasks. We also showcase the advantages of our proposed method in terms of efficiency, scalability, portability, and privacy preservation.
Robots equipped with reinforcement learning (RL) have the potential to learn a wide range of skills solely from a reward signal. However, obtaining a robust and dense reward signal for general manipulation tasks remains a challenge. Existing learning-based approaches require significant data, such as human demonstrations of success and failure, to learn task-specific reward functions. Recently, there is also a growing adoption of large multi-modal foundation models for robotics that can perform visual reasoning in physical contexts and generate coarse robot motions for manipulation tasks. Motivated by this range of capability, in this work, we present Keypoint-based Affordance Guidance for Improvements (KAGI), a method leveraging rewards shaped by vision-language models (VLMs) for autonomous RL. State-of-the-art VLMs have demonstrated impressive reasoning about affordances through keypoints in zero-shot, and we use these to define dense rewards that guide autonomous robotic learning. On real-world manipulation tasks specified by natural language descriptions, KAGI improves the sample efficiency of autonomous RL and enables successful task completion in 30K online fine-tuning steps. Additionally, we demonstrate the robustness of KAGI to reductions in the number of in-domain demonstrations used for pre-training, reaching similar performance in 45K online fine-tuning steps. Project website: https://sites.google.com/view/affordance-guided-rl
Cognitive textual and visual reasoning tasks, including puzzles, series, and analogies, demand the ability to quickly reason, decipher, and evaluate patterns both textually and spatially. Due to extensive training on vast amounts of human-curated data, LLMs and VLMs excel in common-sense reasoning tasks, however still struggle with more complex reasoning that demands deeper cognitive understanding. We introduce NTSEBench, a new dataset designed to evaluate cognitive multi-modal reasoning and problem-solving skills of large models. The dataset contains 2728 multiple-choice questions, accompanied by a total of 4,642 images, categorized into 26 different types. These questions are drawn from the nationwide NTSE examination in India and feature a mix of visual and textual general aptitude challenges, designed to assess intelligence and critical thinking skills beyond mere rote learning. We establish baselines on the dataset using state-of-the-art LLMs and VLMs. To facilitate a comparison between open source and propriety models, we propose four distinct modeling strategies to handle different modalities -- text and images -- in the dataset instances.