Inverse graphics -- the task of inverting an image into physical variables that, when rendered, enable reproduction of the observed scene -- is a fundamental challenge in computer vision and graphics. Successfully disentangling an image into its constituent elements, such as the shape, color, and material properties of the objects of the 3D scene that produced it, requires a comprehensive understanding of the environment. This complexity limits the ability of existing carefully engineered approaches to generalize across domains. Inspired by the zero-shot ability of large language models (LLMs) to generalize to novel contexts, we investigate the possibility of leveraging the broad world knowledge encoded in such models to solve inverse-graphics problems. To this end, we propose the Inverse-Graphics Large Language Model (IG-LLM), an inverse-graphics framework centered around an LLM, that autoregressively decodes a visual embedding into a structured, compositional 3D-scene representation. We incorporate a frozen pre-trained visual encoder and a continuous numeric head to enable end-to-end training. Through our investigation, we demonstrate the potential of LLMs to facilitate inverse graphics through next-token prediction, without the application of image-space supervision. Our analysis enables new possibilities for precise spatial reasoning about images that exploit the visual knowledge of LLMs. We release our code and data at https://ig-llm.is.tue.mpg.de/ to ensure the reproducibility of our investigation and to facilitate future research.
Integration of Large Language Models (LLMs) into visual domain tasks, resulting in visual-LLMs (V-LLMs), has enabled exceptional performance in vision-language tasks, particularly for visual question answering (VQA). However, existing V-LLMs (e.g. BLIP-2, LLaVA) demonstrate weak spatial reasoning and localization awareness. Despite generating highly descriptive and elaborate textual answers, these models fail at simple tasks like distinguishing a left vs right location. In this work, we explore how image-space coordinate based instruction fine-tuning objectives could inject spatial awareness into V-LLMs. We discover optimal coordinate representations, data-efficient instruction fine-tuning objectives, and pseudo-data generation strategies that lead to improved spatial awareness in V-LLMs. Additionally, our resulting model improves VQA across image and video domains, reduces undesired hallucination, and generates better contextual object descriptions. Experiments across 5 vision-language tasks involving 14 different datasets establish the clear performance improvements achieved by our proposed framework.
Combining Large Language Models (LLMs) with external specialized tools (LLMs+tools) is a recent paradigm to solve multimodal tasks such as Visual Question Answering (VQA). While this approach was demonstrated to work well when optimized and evaluated for each individual benchmark, in practice it is crucial for the next generation of real-world AI systems to handle a broad range of multimodal problems. Therefore we pose the VQA problem from a unified perspective and evaluate a single system on a varied suite of VQA tasks including counting, spatial reasoning, OCR-based reasoning, visual pointing, external knowledge, and more. In this setting, we demonstrate that naively applying the LLM+tools approach using the combined set of all tools leads to poor results. This motivates us to introduce HAMMR: HierArchical MultiModal React. We start from a multimodal ReAct-based system and make it hierarchical by enabling our HAMMR agents to call upon other specialized agents. This enhances the compositionality of the LLM+tools approach, which we show to be critical for obtaining high accuracy on generic VQA. Concretely, on our generic VQA suite, HAMMR outperforms the naive LLM+tools approach by 19.5%. Additionally, HAMMR achieves state-of-the-art results on this task, outperforming the generic standalone PaLI-X VQA model by 5.0%.
Flocking is a behavior where multiple agents in a system attempt to stay close to each other while avoiding collision and maintaining a desired formation. This is observed in the natural world and has applications in robotics, including natural disaster search and rescue, wild animal tracking, and perimeter surveillance and patrol. Recently, large language models (LLMs) have displayed an impressive ability to solve various collaboration tasks as individual decision-makers. Solving multi-agent flocking with LLMs would demonstrate their usefulness in situations requiring spatial and decentralized decision-making. Yet, when LLM-powered agents are tasked with implementing multi-agent flocking, they fall short of the desired behavior. After extensive testing, we find that agents with LLMs as individual decision-makers typically opt to converge on the average of their initial positions or diverge from each other. After breaking the problem down, we discover that LLMs cannot understand maintaining a shape or keeping a distance in a meaningful way. Solving multi-agent flocking with LLMs would enhance their ability to understand collaborative spatial reasoning and lay a foundation for addressing more complex multi-agent tasks. This paper discusses the challenges LLMs face in multi-agent flocking and suggests areas for future improvement and research.
Large language models (LLMs) have exhibited impressive performance in language comprehension and various reasoning tasks. However, their abilities in spatial reasoning, a crucial aspect of human cognition, remain relatively unexplored. Human possess a remarkable ability to create mental images of unseen objects and actions through a process known as the Mind's Eye, enabling the imagination of the unseen world. Inspired by this cognitive capacity, we propose Visualization-of-Thought (VoT) prompting. VoT aims to elicit spatial reasoning of LLMs by visualizing their reasoning traces, thereby guiding subsequent reasoning steps. We employed VoT for multi-hop spatial reasoning tasks, including natural language navigation, visual navigation, and visual tiling in 2D grid worlds. Experimental results demonstrated that VoT significantly enhances the spatial reasoning abilities of LLMs. Notably, VoT outperformed existing multimodal large language models (MLLMs) in these tasks. While VoT works surprisingly well on LLMs, the ability to generate mental images to facilitate spatial reasoning resembles the mind's eye process, suggesting its potential viability in MLLMs. Please find the dataset and codes at https://microsoft.github.io/visualization-of-thought
One of the key shortcomings in current text-to-image (T2I) models is their inability to consistently generate images which faithfully follow the spatial relationships specified in the text prompt. In this paper, we offer a comprehensive investigation of this limitation, while also developing datasets and methods that support algorithmic solutions to improve spatial reasoning in T2I models. We find that spatial relationships are under-represented in the image descriptions found in current vision-language datasets. To alleviate this data bottleneck, we create SPRIGHT, the first spatially focused, large-scale dataset, by re-captioning 6 million images from 4 widely used vision datasets and through a 3-fold evaluation and analysis pipeline, show that SPRIGHT improves the proportion of spatial relationships in existing datasets. We show the efficacy of SPRIGHT data by showing that using only $\sim$0.25% of SPRIGHT results in a 22% improvement in generating spatially accurate images while also improving FID and CMMD scores. We also find that training on images containing a larger number of objects leads to substantial improvements in spatial consistency, including state-of-the-art results on T2I-CompBench with a spatial score of 0.2133, by fine-tuning on <500 images. Through a set of controlled experiments and ablations, we document additional findings that could support future work that seeks to understand factors that affect spatial consistency in text-to-image models.
Current state-of-the-art spatial reasoning-enhanced VLMs are trained to excel at spatial visual question answering (VQA). However, we believe that higher-level 3D-aware tasks, such as articulating dynamic scene changes and motion planning, require a fundamental and explicit 3D understanding beyond current spatial VQA datasets. In this work, we present SpatialPIN, a framework designed to enhance the spatial reasoning capabilities of VLMs through prompting and interacting with priors from multiple 3D foundation models in a zero-shot, training-free manner. Extensive experiments demonstrate that our spatial reasoning-imbued VLM performs well on various forms of spatial VQA and can extend to help in various downstream robotics tasks such as pick and stack and trajectory planning.
This paper shows that text-only Language Models (LM) can learn to ground spatial relations like "left of" or "below" if they are provided with explicit location information of objects and they are properly trained to leverage those locations. We perform experiments on a verbalized version of the Visual Spatial Reasoning (VSR) dataset, where images are coupled with textual statements which contain real or fake spatial relations between two objects of the image. We verbalize the images using an off-the-shelf object detector, adding location tokens to every object label to represent their bounding boxes in textual form. Given the small size of VSR, we do not observe any improvement when using locations, but pretraining the LM over a synthetic dataset automatically derived by us improves results significantly when using location tokens. We thus show that locations allow LMs to ground spatial relations, with our text-only LMs outperforming Vision-and-Language Models and setting the new state-of-the-art for the VSR dataset. Our analysis show that our text-only LMs can generalize beyond the relations seen in the synthetic dataset to some extent, learning also more useful information than that encoded in the spatial rules we used to create the synthetic dataset itself.
Event-based moving object detection is a challenging task, where static background and moving object are mixed together. Typically, existing methods mainly align the background events to the same spatial coordinate system via motion compensation to distinguish the moving object. However, they neglect the potential spatial tailing effect of moving object events caused by excessive motion, which may affect the structure integrity of the extracted moving object. We discover that the moving object has a complete columnar structure in the point cloud composed of motion-compensated events along the timestamp. Motivated by this, we propose a novel joint spatio-temporal reasoning method for event-based moving object detection. Specifically, we first compensate the motion of background events using inertial measurement unit. In spatial reasoning stage, we project the compensated events into the same image coordinate, discretize the timestamp of events to obtain a time image that can reflect the motion confidence, and further segment the moving object through adaptive threshold on the time image. In temporal reasoning stage, we construct the events into a point cloud along timestamp, and use RANSAC algorithm to extract the columnar shape in the cloud for peeling off the background. Finally, we fuse the results from the two reasoning stages to extract the final moving object region. This joint spatio-temporal reasoning framework can effectively detect the moving object from motion confidence and geometric structure. Moreover, we conduct extensive experiments on various datasets to verify that the proposed method can improve the moving object detection accuracy by 13\%.
Diffusion-driven text-to-image (T2I) generation has achieved remarkable advancements. To further improve T2I models' capability in numerical and spatial reasoning, the layout is employed as an intermedium to bridge large language models and layout-based diffusion models. However, these methods still struggle with generating images from textural prompts with multiple objects and complicated spatial relationships. To tackle this challenge, we introduce a divide-and-conquer approach which decouples the T2I generation task into simple subtasks. Our approach divides the layout prediction stage into numerical & spatial reasoning and bounding box prediction. Then, the layout-to-image generation stage is conducted in an iterative manner to reconstruct objects from easy ones to difficult ones. We conduct experiments on the HRS and NSR-1K benchmarks and our approach outperforms previous state-of-the-art models with notable margins. In addition, visual results demonstrate that our approach significantly improves the controllability and consistency in generating multiple objects from complex textural prompts.
Highlighting particularly relevant regions of an image can improve the performance of vision-language models (VLMs) on various vision-language (VL) tasks by guiding the model to attend more closely to these regions of interest. For example, VLMs can be given a "visual prompt", where visual markers such as bounding boxes delineate key image regions. However, current VLMs that can incorporate visual guidance are either proprietary and expensive or require costly training on curated data that includes visual prompts. We introduce Contrastive Region Guidance (CRG), a training-free guidance method that enables open-source VLMs to respond to visual prompts. CRG contrasts model outputs produced with and without visual prompts, factoring out biases revealed by the model when answering without the information required to produce a correct answer (i.e., the model's prior). CRG achieves substantial improvements in a wide variety of VL tasks: When region annotations are provided, CRG increases absolute accuracy by up to 11.1% on ViP-Bench, a collection of six diverse region-based tasks such as recognition, math, and object relationship reasoning. We also show CRG's applicability to spatial reasoning, with 10% improvement on What'sUp, as well as to compositional generalization -- improving accuracy by 11.5% and 7.5% on two challenging splits from SugarCrepe -- and to image-text alignment for generated images, where we improve by up to 8.4 AUROC and 6.8 F1 points on SeeTRUE. When reference regions are absent, CRG allows us to re-rank proposed regions in referring expression comprehension and phrase grounding benchmarks like RefCOCO/+/g and Flickr30K Entities, with an average gain of 3.2% in accuracy. Our analysis explores alternative masking strategies for CRG, quantifies CRG's probability shift, and evaluates the role of region guidance strength, empirically validating CRG's design choices.
This study evaluates three state-of-the-art MLLMs -- GPT-4V, Gemini Pro, and the open-source model IDEFICS -- on the compositional natural language vision reasoning task NLVR. Given a human-written sentence paired with a synthetic image, this task requires the model to determine the truth value of the sentence with respect to the image. Despite the strong performance demonstrated by these models, we observe they perform poorly on NLVR, which was constructed to require compositional and spatial reasoning, and to be robust for semantic and systematic biases.
Recent advancements in large language models (LLMs) have revealed their potential for achieving autonomous agents possessing human-level intelligence. However, existing benchmarks for evaluating LLM Agents either use static datasets, potentially leading to data leakage or focus only on single-agent scenarios, overlooking the complexities of multi-agent interactions. There is a lack of a benchmark that evaluates the diverse capabilities of LLM agents in multi-agent, dynamic environments. To this end, we introduce LLMArena, a novel and easily extensible framework for evaluating the diverse capabilities of LLM in multi-agent dynamic environments. LLMArena encompasses seven distinct gaming environments, employing Trueskill scoring to assess crucial abilities in LLM agents, including spatial reasoning, strategic planning, numerical reasoning, risk assessment, communication, opponent modeling, and team collaboration. We conduct an extensive experiment and human evaluation among different sizes and types of LLMs, showing that LLMs still have a significant journey ahead in their development towards becoming fully autonomous agents, especially in opponent modeling and team collaboration. We hope LLMArena could guide future research towards enhancing these capabilities in LLMs, ultimately leading to more sophisticated and practical applications in dynamic, multi-agent settings. The code and data will be available.
A primary hurdle of autonomous driving in urban environments is understanding complex and long-tail scenarios, such as challenging road conditions and delicate human behaviors. We introduce DriveVLM, an autonomous driving system leveraging Vision-Language Models (VLMs) for enhanced scene understanding and planning capabilities. DriveVLM integrates a unique combination of reasoning modules for scene description, scene analysis, and hierarchical planning. Furthermore, recognizing the limitations of VLMs in spatial reasoning and heavy computational requirements, we propose DriveVLM-Dual, a hybrid system that synergizes the strengths of DriveVLM with the traditional autonomous driving pipeline. Experiments on both the nuScenes dataset and our SUP-AD dataset demonstrate the efficacy of DriveVLM and DriveVLM-Dual in handling complex and unpredictable driving conditions. Finally, we deploy the DriveVLM-Dual on a production vehicle, verifying it is effective in real-world autonomous driving environments.
Vision language models (VLMs) have shown impressive capabilities across a variety of tasks, from logical reasoning to visual understanding. This opens the door to richer interaction with the world, for example robotic control. However, VLMs produce only textual outputs, while robotic control and other spatial tasks require outputting continuous coordinates, actions, or trajectories. How can we enable VLMs to handle such settings without fine-tuning on task-specific data? In this paper, we propose a novel visual prompting approach for VLMs that we call Prompting with Iterative Visual Optimization (PIVOT), which casts tasks as iterative visual question answering. In each iteration, the image is annotated with a visual representation of proposals that the VLM can refer to (e.g., candidate robot actions, localizations, or trajectories). The VLM then selects the best ones for the task. These proposals are iteratively refined, allowing the VLM to eventually zero in on the best available answer. We investigate PIVOT on real-world robotic navigation, real-world manipulation from images, instruction following in simulation, and additional spatial inference tasks such as localization. We find, perhaps surprisingly, that our approach enables zero-shot control of robotic systems without any robot training data, navigation in a variety of environments, and other capabilities. Although current performance is far from perfect, our work highlights potentials and limitations of this new regime and shows a promising approach for Internet-Scale VLMs in robotic and spatial reasoning domains. Website: pivot-prompt.github.io and HuggingFace: https://huggingface.co/spaces/pivot-prompt/pivot-prompt-demo.
The advent of foundation models has revolutionized the fields of natural language processing and computer vision, paving the way for their application in autonomous driving (AD). This survey presents a comprehensive review of more than 40 research papers, demonstrating the role of foundation models in enhancing AD. Large language models contribute to planning and simulation in AD, particularly through their proficiency in reasoning, code generation and translation. In parallel, vision foundation models are increasingly adapted for critical tasks such as 3D object detection and tracking, as well as creating realistic driving scenarios for simulation and testing. Multi-modal foundation models, integrating diverse inputs, exhibit exceptional visual understanding and spatial reasoning, crucial for end-to-end AD. This survey not only provides a structured taxonomy, categorizing foundation models based on their modalities and functionalities within the AD domain but also delves into the methods employed in current research. It identifies the gaps between existing foundation models and cutting-edge AD approaches, thereby charting future research directions and proposing a roadmap for bridging these gaps.
Large Vision-Language Models (VLMs) have demonstrated impressive performance on complex tasks involving visual input with natural language instructions. However, it remains unclear to what extent capabilities on natural images transfer to Earth observation (EO) data, which are predominantly satellite and aerial images less common in VLM training data. In this work, we propose a comprehensive benchmark to gauge the progress of VLMs toward being useful tools for EO data by assessing their abilities on scene understanding, localization and counting, and change detection tasks. Motivated by real-world applications, our benchmark includes scenarios like urban monitoring, disaster relief, land use, and conservation. We discover that, although state-of-the-art VLMs like GPT-4V possess extensive world knowledge that leads to strong performance on open-ended tasks like location understanding and image captioning, their poor spatial reasoning limits usefulness on object localization and counting tasks. Our benchmark will be made publicly available at https://vleo.danielz.ch/ and on Hugging Face at https://huggingface.co/collections/mit-ei/vleo-benchmark-datasets-65b789b0466555489cce0d70 for easy model evaluation.
Understanding and reasoning about spatial relationships is a fundamental capability for Visual Question Answering (VQA) and robotics. While Vision Language Models (VLM) have demonstrated remarkable performance in certain VQA benchmarks, they still lack capabilities in 3D spatial reasoning, such as recognizing quantitative relationships of physical objects like distances or size differences. We hypothesize that VLMs' limited spatial reasoning capability is due to the lack of 3D spatial knowledge in training data and aim to solve this problem by training VLMs with Internet-scale spatial reasoning data. To this end, we present a system to facilitate this approach. We first develop an automatic 3D spatial VQA data generation framework that scales up to 2 billion VQA examples on 10 million real-world images. We then investigate various factors in the training recipe, including data quality, training pipeline, and VLM architecture. Our work features the first internet-scale 3D spatial reasoning dataset in metric space. By training a VLM on such data, we significantly enhance its ability on both qualitative and quantitative spatial VQA. Finally, we demonstrate that this VLM unlocks novel downstream applications in chain-of-thought spatial reasoning and robotics due to its quantitative estimation capability. Project website: https://spatial-vlm.github.io/
Fully understanding a complex high-resolution satellite or aerial imagery scene often requires spatial reasoning over a broad relevant context. The human object recognition system is able to understand object in a scene over a long-range relevant context. For example, if a human observes an aerial scene that shows sections of road broken up by tree canopy, then they will be unlikely to conclude that the road has actually been broken up into disjoint pieces by trees and instead think that the canopy of nearby trees is occluding the road. However, there is limited research being conducted to understand long-range context understanding of modern machine learning models. In this work we propose a road segmentation benchmark dataset, Chesapeake Roads Spatial Context (RSC), for evaluating the spatial long-range context understanding of geospatial machine learning models and show how commonly used semantic segmentation models can fail at this task. For example, we show that a U-Net trained to segment roads from background in aerial imagery achieves an 84% recall on unoccluded roads, but just 63.5% recall on roads covered by tree canopy despite being trained to model both the same way. We further analyze how the performance of models changes as the relevant context for a decision (unoccluded roads in our case) varies in distance. We release the code to reproduce our experiments and dataset of imagery and masks to encourage future research in this direction -- https://github.com/isaaccorley/ChesapeakeRSC.
We present a benchmark for assessing the capability of Large Language Models (LLMs) to discern intercardinal directions between geographic locations and apply it to three prominent LLMs: GPT-3.5, GPT-4, and Llama-2. This benchmark specifically evaluates whether LLMs exhibit a hierarchical spatial bias similar to humans, where judgments about individual locations' spatial relationships are influenced by the perceived relationships of the larger groups that contain them. To investigate this, we formulated 14 questions focusing on well-known American cities. Seven questions were designed to challenge the LLMs with scenarios potentially influenced by the orientation of larger geographical units, such as states or countries, while the remaining seven targeted locations were less susceptible to such hierarchical categorization. Among the tested models, GPT-4 exhibited superior performance with 55 percent accuracy, followed by GPT-3.5 at 47 percent, and Llama-2 at 45 percent. The models showed significantly reduced accuracy on tasks with suspected hierarchical bias. For example, GPT-4's accuracy dropped to 33 percent on these tasks, compared to 86 percent on others. However, the models identified the nearest cardinal direction in most cases, reflecting their associative learning mechanism, thereby embodying human-like misconceptions. We discuss avenues for improving the spatial reasoning capabilities of LLMs.