The Large Visual-Language Models (LVLMs) have significantly advanced image understanding. Their comprehension and reasoning capabilities enable promising applications in autonomous driving scenarios. However, existing research typically focuses on front-view perspectives and partial objects within scenes, struggling to achieve comprehensive scene understanding. Meanwhile, existing LVLMs suffer from the lack of mapping relationship between 2D and 3D and insufficient integration of 3D object localization and instruction understanding. To tackle these limitations, we first introduce NuInteract, a large-scale dataset with over 1.5M multi-view image language pairs spanning dense scene captions and diverse interactive tasks. Furthermore, we propose DriveMonkey, a simple yet effective framework that seamlessly integrates LVLMs with a spatial processor using a series of learnable queries. The spatial processor, designed as a plug-and-play component, can be initialized with pre-trained 3D detectors to improve 3D perception. Our experiments show that DriveMonkey outperforms general LVLMs, especially achieving a 9.86% notable improvement on the 3D visual grounding task. The dataset and code will be released at https://github.com/zc-zhao/DriveMonkey.
This report presents our semantic segmentation framework developed by team ACVLAB for the ICRA 2025 GOOSE 2D Semantic Segmentation Challenge, which focuses on parsing outdoor scenes into nine semantic categories under real-world conditions. Our method integrates a Swin Transformer backbone enhanced with Rotary Position Embedding (RoPE) for improved spatial generalization, alongside a Color Shift Estimation-and-Correction module designed to compensate for illumination inconsistencies in natural environments. To further improve training stability, we adopt a quantile-based denoising strategy that downweights the top 2.5\% of highest-error pixels, treating them as noise and suppressing their influence during optimization. Evaluated on the official GOOSE test set, our approach achieved a mean Intersection over Union (mIoU) of 0.848, demonstrating the effectiveness of combining color correction, positional encoding, and error-aware denoising in robust semantic segmentation.
In autonomous driving, thermal image semantic segmentation has emerged as a critical research area, owing to its ability to provide robust scene understanding under adverse visual conditions. In particular, unsupervised domain adaptation (UDA) for thermal image segmentation can be an efficient solution to address the lack of labeled thermal datasets. Nevertheless, since these methods do not effectively utilize the complementary information between RGB and thermal images, they significantly decrease performance during domain adaptation. In this paper, we present a comprehensive study on cross-spectral UDA for thermal image semantic segmentation. We first propose a novel masked mutual learning strategy that promotes complementary information exchange by selectively transferring results between each spectral model while masking out uncertain regions. Additionally, we introduce a novel prototypical self-supervised loss designed to enhance the performance of the thermal segmentation model in nighttime scenarios. This approach addresses the limitations of RGB pre-trained networks, which cannot effectively transfer knowledge under low illumination due to the inherent constraints of RGB sensors. In experiments, our method achieves higher performance over previous UDA methods and comparable performance to state-of-the-art supervised methods.
Traffic accident prediction and detection are critical for enhancing road safety,and vision-based traffic accident anticipation (Vision-TAA) has emerged as a promising approach in the era of deep learning.This paper reviews 147 recent studies,focusing on the application of supervised,unsupervised,and hybrid deep learning models for accident prediction,alongside the use of real-world and synthetic datasets.Current methodologies are categorized into four key approaches: image and video feature-based prediction, spatiotemporal feature-based prediction, scene understanding,and multimodal data fusion.While these methods demonstrate significant potential,challenges such as data scarcity,limited generalization to complex scenarios,and real-time performance constraints remain prevalent. This review highlights opportunities for future research,including the integration of multimodal data fusion, self-supervised learning,and Transformer-based architectures to enhance prediction accuracy and scalability.By synthesizing existing advancements and identifying critical gaps, this paper provides a foundational reference for developing robust and adaptive Vision-TAA systems,contributing to road safety and traffic management.
In this paper, we present Optimized Prompt-based Unified System (OPUS), a framework that utilizes a Large Language Model (LLM) to control Pan-Tilt-Zoom (PTZ) cameras, providing contextual understanding of natural environments. To achieve this goal, the OPUS system improves cost-effectiveness by generating keywords from a high-level camera control API and transferring knowledge from larger closed-source language models to smaller ones through Supervised Fine-Tuning (SFT) on synthetic data. This enables efficient edge deployment while maintaining performance comparable to larger models like GPT-4. OPUS enhances environmental awareness by converting data from multiple cameras into textual descriptions for language models, eliminating the need for specialized sensory tokens. In benchmark testing, our approach significantly outperformed both traditional language model techniques and more complex prompting methods, achieving a 35% improvement over advanced techniques and a 20% higher task accuracy compared to closed-source models like Gemini Pro. The system demonstrates OPUS's capability to simplify PTZ camera operations through an intuitive natural language interface. This approach eliminates the need for explicit programming and provides a conversational method for interacting with camera systems, representing a significant advancement in how users can control and utilize PTZ camera technology.
Autonomous vehicle perception systems have traditionally relied on costly LiDAR sensors to generate precise environmental representations. In this paper, we propose a camera-only perception framework that produces Bird's Eye View (BEV) maps by extending the Lift-Splat-Shoot architecture. Our method combines YOLOv11-based object detection with DepthAnythingV2 monocular depth estimation across multi-camera inputs to achieve comprehensive 360-degree scene understanding. We evaluate our approach on the OpenLane-V2 and NuScenes datasets, achieving up to 85% road segmentation accuracy and 85-90% vehicle detection rates when compared against LiDAR ground truth, with average positional errors limited to 1.2 meters. These results highlight the potential of deep learning to extract rich spatial information using only camera inputs, enabling cost-efficient autonomous navigation without sacrificing accuracy.
We present the challenging task of automatically creating a high-level Wikipedia-style article that aggregates information from multiple diverse videos about real-world events, such as natural disasters or political elections. Videos are intuitive sources for retrieval-augmented generation (RAG), but most contemporary RAG workflows focus heavily on text and existing methods for video-based summarization focus on low-level scene understanding rather than high-level event semantics. To close this gap, we introduce WikiVideo, a benchmark consisting of expert-written articles and densely annotated videos that provide evidence for articles' claims, facilitating the integration of video into RAG pipelines and enabling the creation of in-depth content that is grounded in multimodal sources. We further propose Collaborative Article Generation (CAG), a novel interactive method for article creation from multiple videos. CAG leverages an iterative interaction between an r1-style reasoning model and a VideoLLM to draw higher level inferences about the target event than is possible with VideoLLMs alone, which fixate on low-level visual features. We benchmark state-of-the-art VideoLLMs and CAG in both oracle retrieval and RAG settings and find that CAG consistently outperforms alternative methods, while suggesting intriguing avenues for future work.
We propose a new task to benchmark human-in-scene understanding for embodied agents: Human-In-Scene Question Answering (HIS-QA). Given a human motion within a 3D scene, HIS-QA requires the agent to comprehend human states and behaviors, reason about its surrounding environment, and answer human-related questions within the scene. To support this new task, we present HIS-Bench, a multimodal benchmark that systematically evaluates HIS understanding across a broad spectrum, from basic perception to commonsense reasoning and planning. Our evaluation of various vision-language models on HIS-Bench reveals significant limitations in their ability to handle HIS-QA tasks. To this end, we propose HIS-GPT, the first foundation model for HIS understanding. HIS-GPT integrates 3D scene context and human motion dynamics into large language models while incorporating specialized mechanisms to capture human-scene interactions. Extensive experiments demonstrate that HIS-GPT sets a new state-of-the-art on HIS-QA tasks. We hope this work inspires future research on human behavior analysis in 3D scenes, advancing embodied AI and world models.
Large-scale vision-language models demonstrate strong multimodal alignment and generalization across diverse tasks. Among them, CLIP stands out as one of the most successful approaches. In this work, we extend the application of CLIP to sound source localization, proposing a self-supervised method operates without explicit text input. We introduce a framework that maps audios into tokens compatible with CLIP's text encoder, producing audio-driven embeddings. These embeddings are used to generate sounding region masks, from which visual features are extracted and aligned with the audio embeddings through a contrastive audio-visual correspondence objective. Our findings show that alignment knowledge of pre-trained multimodal foundation model enables our method to generate more complete and compact localization for sounding objects. We further propose an LLM-guided extension that distills object-aware audio-visual scene understanding into the model during training to enhance alignment. Extensive experiments across five diverse tasks demonstrate that our method, in all variants, outperforms state-of-the-art approaches and achieves strong generalization in zero-shot settings.
In this paper, we propose PADriver, a novel closed-loop framework for personalized autonomous driving (PAD). Built upon Multi-modal Large Language Model (MLLM), PADriver takes streaming frames and personalized textual prompts as inputs. It autoaggressively performs scene understanding, danger level estimation and action decision. The predicted danger level reflects the risk of the potential action and provides an explicit reference for the final action, which corresponds to the preset personalized prompt. Moreover, we construct a closed-loop benchmark named PAD-Highway based on Highway-Env simulator to comprehensively evaluate the decision performance under traffic rules. The dataset contains 250 hours videos with high-quality annotation to facilitate the development of PAD behavior analysis. Experimental results on the constructed benchmark show that PADriver outperforms state-of-the-art approaches on different evaluation metrics, and enables various driving modes.
CLIP has emerged as a powerful multimodal model capable of connecting images and text through joint embeddings, but to what extent does it "see" the same way humans do - especially when interpreting artworks? In this paper, we investigate CLIP's ability to extract high-level semantic and stylistic information from paintings, including both human-created and AI-generated imagery. We evaluate its perception across multiple dimensions: content, scene understanding, artistic style, historical period, and the presence of visual deformations or artifacts. By designing targeted probing tasks and comparing CLIP's responses to human annotations and expert benchmarks, we explore its alignment with human perceptual and contextual understanding. Our findings reveal both strengths and limitations in CLIP's visual representations, particularly in relation to aesthetic cues and artistic intent. We further discuss the implications of these insights for using CLIP as a guidance mechanism during generative processes, such as style transfer or prompt-based image synthesis. Our work highlights the need for deeper interpretability in multimodal systems, especially when applied to creative domains where nuance and subjectivity play a central role.
Traffic safety remains a critical global challenge, with traditional Advanced Driver-Assistance Systems (ADAS) often struggling in dynamic real-world scenarios due to fragmented sensor processing and susceptibility to adversarial conditions. This paper reviews the transformative potential of Multimodal Large Language Models (MLLMs) in addressing these limitations by integrating cross-modal data such as visual, spatial, and environmental inputs to enable holistic scene understanding. Through a comprehensive analysis of MLLM-based approaches, we highlight their capabilities in enhancing perception, decision-making, and adversarial robustness, while also examining the role of key datasets (e.g., KITTI, DRAMA, ML4RoadSafety) in advancing research. Furthermore, we outline future directions, including real-time edge deployment, causality-driven reasoning, and human-AI collaboration. By positioning MLLMs as a cornerstone for next-generation traffic safety systems, this review underscores their potential to revolutionize the field, offering scalable, context-aware solutions that proactively mitigate risks and improve overall road safety.
We investigate the ability of Vision Language Models (VLMs) to perform visual perspective taking using a novel set of visual tasks inspired by established human tests. Our approach leverages carefully controlled scenes, in which a single humanoid minifigure is paired with a single object. By systematically varying spatial configurations - such as object position relative to the humanoid minifigure and the humanoid minifigure's orientation - and using both bird's-eye and surface-level views, we created 144 unique visual tasks. Each visual task is paired with a series of 7 diagnostic questions designed to assess three levels of visual cognition: scene understanding, spatial reasoning, and visual perspective taking. Our evaluation of several state-of-the-art models, including GPT-4-Turbo, GPT-4o, Llama-3.2-11B-Vision-Instruct, and variants of Claude Sonnet, reveals that while they excel in scene understanding, the performance declines significantly on spatial reasoning and further deteriorates on perspective-taking. Our analysis suggests a gap between surface-level object recognition and the deeper spatial and perspective reasoning required for complex visual tasks, pointing to the need for integrating explicit geometric representations and tailored training protocols in future VLM development.
Image segmentation is a powerful computer vision technique for scene understanding. However, real-world deployment is stymied by the need for high-quality, meticulously labeled datasets. Synthetic data provides high-quality labels while reducing the need for manual data collection and annotation. However, deep neural networks trained on synthetic data often face the Syn2Real problem, leading to poor performance in real-world deployments. To mitigate the aforementioned gap in image segmentation, we propose RAFT, a novel framework for adapting image segmentation models using minimal labeled real-world data through data and feature augmentations, as well as active learning. To validate RAFT, we perform experiments on the synthetic-to-real "SYNTHIA->Cityscapes" and "GTAV->Cityscapes" benchmarks. We managed to surpass the previous state of the art, HALO. SYNTHIA->Cityscapes experiences an improvement in mIoU* upon domain adaptation of 2.1%/79.9%, and GTAV->Cityscapes experiences a 0.4%/78.2% improvement in mIoU. Furthermore, we test our approach on the real-to-real benchmark of "Cityscapes->ACDC", and again surpass HALO, with a gain in mIoU upon adaptation of 1.3%/73.2%. Finally, we examine the effect of the allocated annotation budget and various components of RAFT upon the final transfer mIoU.
Open-vocabulary 3D scene understanding is pivotal for enhancing physical intelligence, as it enables embodied agents to interpret and interact dynamically within real-world environments. This paper introduces MPEC, a novel Masked Point-Entity Contrastive learning method for open-vocabulary 3D semantic segmentation that leverages both 3D entity-language alignment and point-entity consistency across different point cloud views to foster entity-specific feature representations. Our method improves semantic discrimination and enhances the differentiation of unique instances, achieving state-of-the-art results on ScanNet for open-vocabulary 3D semantic segmentation and demonstrating superior zero-shot scene understanding capabilities. Extensive fine-tuning experiments on 8 datasets, spanning from low-level perception to high-level reasoning tasks, showcase the potential of learned 3D features, driving consistent performance gains across varied 3D scene understanding tasks. Project website: https://mpec-3d.github.io/
Self-supervised learning has transformed 2D computer vision by enabling models trained on large, unannotated datasets to provide versatile off-the-shelf features that perform similarly to models trained with labels. However, in 3D scene understanding, self-supervised methods are typically only used as a weight initialization step for task-specific fine-tuning, limiting their utility for general-purpose feature extraction. This paper addresses this shortcoming by proposing a robust evaluation protocol specifically designed to assess the quality of self-supervised features for 3D scene understanding. Our protocol uses multi-resolution feature sampling of hierarchical models to create rich point-level representations that capture the semantic capabilities of the model and, hence, are suitable for evaluation with linear probing and nearest-neighbor methods. Furthermore, we introduce the first self-supervised model that performs similarly to supervised models when only off-the-shelf features are used in a linear probing setup. In particular, our model is trained natively in 3D with a novel self-supervised approach based on a Masked Scene Modeling objective, which reconstructs deep features of masked patches in a bottom-up manner and is specifically tailored to hierarchical 3D models. Our experiments not only demonstrate that our method achieves competitive performance to supervised models, but also surpasses existing self-supervised approaches by a large margin. The model and training code can be found at our Github repository (https://github.com/phermosilla/msm).
Driving scene understanding is a critical real-world problem that involves interpreting and associating various elements of a driving environment, such as vehicles, pedestrians, and traffic signals. Despite advancements in autonomous driving, traditional pipelines rely on deterministic models that fail to capture the probabilistic nature and inherent uncertainty of real-world driving. To address this, we propose PRIMEDrive-CoT, a novel uncertainty-aware model for object interaction and Chain-of-Thought (CoT) reasoning in driving scenarios. In particular, our approach combines LiDAR-based 3D object detection with multi-view RGB references to ensure interpretable and reliable scene understanding. Uncertainty and risk assessment, along with object interactions, are modelled using Bayesian Graph Neural Networks (BGNNs) for probabilistic reasoning under ambiguous conditions. Interpretable decisions are facilitated through CoT reasoning, leveraging object dynamics and contextual cues, while Grad-CAM visualizations highlight attention regions. Extensive evaluations on the DriveCoT dataset demonstrate that PRIMEDrive-CoT outperforms state-of-the-art CoT and risk-aware models.
Wireless signals are integral to modern society, enabling both communication and increasingly, environmental sensing. While various propagation models exist, ranging from empirical methods to full-wave simulations, the phenomenon of electromagnetic diffraction is often treated as a secondary effect or a correction factor. This paper positions diffraction as a fundamentally important and underutilized mechanism that is rich with information about the physical environment. Specifically, diffraction-inducing elements generate distinct signatures that are rich with information about their underlying properties such as their geometries. We then argue that by understanding and exploiting these relationships, diffraction can be harnessed strategically. We introduce a general optimization framework to formalize this concept, illustrating how diffraction can be leveraged for both inverse problems (sensing scene details such as object geometries from measured fields) and forward problems (shaping RF fields for communication objectives by configuring diffracting elements). Focusing primarily on edge diffraction and Keller's Geometrical Theory of Diffraction (GTD), we discuss specific applications in RF sensing for scene understanding and in communications for RF field programming, drawing upon recent work. Overall, this paper lays out a vision for systematically incorporating diffraction into the design and operation of future wireless systems, paving the way for enhanced sensing capabilities and more robust communication strategies.
Object pose estimation enables a variety of tasks in computer vision and robotics, including scene understanding and robotic grasping. The complexity of a pose estimation task depends on the unknown variables related to the target object. While instance-level methods already excel for opaque and Lambertian objects, category-level and open-set methods, where texture, shape, and size are partially or entirely unknown, still struggle with these basic material properties. Since texture is unknown in these scenarios, it cannot be used for disambiguating object symmetries, another core challenge of 6D object pose estimation. The complexity of estimating 6D poses with such a manifold of unknowns led to various datasets, accuracy metrics, and algorithmic solutions. This paper compares datasets, accuracy metrics, and algorithms for solving 6D pose estimation on the category-level. Based on this comparison, we analyze how to bridge category-level and open-set object pose estimation to reach generalization and provide actionable recommendations.
We present an embodied robotic system with an LLM-driven agent-orchestration architecture for autonomous household object management. The system integrates memory-augmented task planning, enabling robots to execute high-level user commands while tracking past actions. It employs three specialized agents: a routing agent, a task planning agent, and a knowledge base agent, each powered by task-specific LLMs. By leveraging in-context learning, our system avoids the need for explicit model training. RAG enables the system to retrieve context from past interactions, enhancing long-term object tracking. A combination of Grounded SAM and LLaMa3.2-Vision provides robust object detection, facilitating semantic scene understanding for task planning. Evaluation across three household scenarios demonstrates high task planning accuracy and an improvement in memory recall due to RAG. Specifically, Qwen2.5 yields best performance for specialized agents, while LLaMA3.1 excels in routing tasks. The source code is available at: https://github.com/marc1198/chat-hsr.