Visual Representations inside the Language Model

Despite interpretability work analyzing VIT encoders and transformer activations, we don't yet understand why Multimodal Language Models (MLMs) struggle on perception-heavy tasks. We offer an under-studied perspective by examining how popular MLMs (LLaVA-OneVision, Qwen2.5-VL, and Llama-3-LLaVA-NeXT) process their visual key-value tokens. We first study the flow of visual information through the language model, finding that image value tokens encode sufficient information to perform several perception-heavy tasks zero-shot: segmentation, semantic correspondence, temporal correspondence, and referring expression detection. We find that while the language model does augment the visual information received from the projection of input visual encodings-which we reveal correlates with overall MLM perception capability-it contains less visual information on several tasks than the equivalent visual encoder (SigLIP) that has not undergone MLM finetuning. Further, we find that the visual information corresponding to input-agnostic image key tokens in later layers of language models contains artifacts which reduce perception capability of the overall MLM. Next, we discuss controlling visual information in the language model, showing that adding a text prefix to the image input improves perception capabilities of visual representations. Finally, we reveal that if language models were able to better control their visual information, their perception would significantly improve; e.g., in 33.3% of Art Style questions in the BLINK benchmark, perception information present in the language model is not surfaced to the output! Our findings reveal insights into the role of key-value tokens in multimodal systems, paving the way for deeper mechanistic interpretability of MLMs and suggesting new directions for training their visual encoder and language model components.

When Tracking Fails: Analyzing Failure Modes of SAM2 for Point-Based Tracking in Surgical Videos

Video object segmentation (VOS) models such as SAM2 offer promising zero-shot tracking capabilities for surgical videos using minimal user input. Among the available input types, point-based tracking offers an efficient and low-cost alternative, yet its reliability and failure cases in complex surgical environments are not well understood. In this work, we systematically analyze the failure modes of point-based tracking in laparoscopic cholecystectomy videos. Focusing on three surgical targets, the gallbladder, grasper, and L-hook electrocautery, we compare the performance of point-based tracking with segmentation mask initialization. Our results show that point-based tracking is competitive for surgical tools but consistently underperforms for anatomical targets, where tissue similarity and ambiguous boundaries lead to failure. Through qualitative analysis, we reveal key factors influencing tracking outcomes and provide several actionable recommendations for selecting and placing tracking points to improve performance in surgical video analysis.

Solar PV Installation Potential Assessment on Building Facades Based on Vision and Language Foundation Models

Building facades represent a significant untapped resource for solar energy generation in dense urban environments, yet assessing their photovoltaic (PV) potential remains challenging due to complex geometries and semantic com ponents. This study introduces SF-SPA (Semantic Facade Solar-PV Assessment), an automated framework that transforms street-view photographs into quantitative PV deployment assessments. The approach combines com puter vision and artificial intelligence techniques to address three key challenges: perspective distortion correction, semantic understanding of facade elements, and spatial reasoning for PV layout optimization. Our four-stage pipeline processes images through geometric rectification, zero-shot semantic segmentation, Large Language Model (LLM) guided spatial reasoning, and energy simulation. Validation across 80 buildings in four countries demonstrates ro bust performance with mean area estimation errors of 6.2% ± 2.8% compared to expert annotations. The auto mated assessment requires approximately 100 seconds per building, a substantial gain in efficiency over manual methods. Simulated energy yield predictions confirm the method's reliability and applicability for regional poten tial studies, urban energy planning, and building-integrated photovoltaic (BIPV) deployment. Code is available at: https:github.com/CodeAXu/Solar-PV-Installation

KAIROS: Unified Training for Universal Non-Autoregressive Time Series Forecasting

In the World Wide Web, reliable time series forecasts provide the forward-looking signals that drive resource planning, cache placement, and anomaly response, enabling platforms to operate efficiently as user behavior and content distributions evolve. Compared with other domains, time series forecasting for Web applications requires much faster responsiveness to support real-time decision making. We present KAIROS, a non-autoregressive time series forecasting framework that directly models segment-level multi-peak distributions. Unlike autoregressive approaches, KAIROS avoids error accumulation and achieves just-in-time inference, while improving over existing non-autoregressive models that collapse to over-smoothed predictions. Trained on the large-scale corpus, KAIROS demonstrates strong zero-shot generalization on six widely used benchmarks, delivering forecasting performance comparable to state-of-the-art foundation models with similar scale, at a fraction of their inference cost. Beyond empirical results, KAIROS highlights the importance of non-autoregressive design as a scalable paradigm for foundation models in time series.

RefAM: Attention Magnets for Zero-Shot Referral Segmentation

Most existing approaches to referring segmentation achieve strong performance only through fine-tuning or by composing multiple pre-trained models, often at the cost of additional training and architectural modifications. Meanwhile, large-scale generative diffusion models encode rich semantic information, making them attractive as general-purpose feature extractors. In this work, we introduce a new method that directly exploits features, attention scores, from diffusion transformers for downstream tasks, requiring neither architectural modifications nor additional training. To systematically evaluate these features, we extend benchmarks with vision-language grounding tasks spanning both images and videos. Our key insight is that stop words act as attention magnets: they accumulate surplus attention and can be filtered to reduce noise. Moreover, we identify global attention sinks (GAS) emerging in deeper layers and show that they can be safely suppressed or redirected onto auxiliary tokens, leading to sharper and more accurate grounding maps. We further propose an attention redistribution strategy, where appended stop words partition background activations into smaller clusters, yielding sharper and more localized heatmaps. Building on these findings, we develop RefAM, a simple training-free grounding framework that combines cross-attention maps, GAS handling, and redistribution. Across zero-shot referring image and video segmentation benchmarks, our approach consistently outperforms prior methods, establishing a new state of the art without fine-tuning or additional components.

TASAM: Terrain-and-Aware Segment Anything Model for Temporal-Scale Remote Sensing Segmentation

Segment Anything Model (SAM) has demonstrated impressive zero-shot segmentation capabilities across natural image domains, but it struggles to generalize to the unique challenges of remote sensing data, such as complex terrain, multi-scale objects, and temporal dynamics. In this paper, we introduce TASAM, a terrain and temporally-aware extension of SAM designed specifically for high-resolution remote sensing image segmentation. TASAM integrates three lightweight yet effective modules: a terrain-aware adapter that injects elevation priors, a temporal prompt generator that captures land-cover changes over time, and a multi-scale fusion strategy that enhances fine-grained object delineation. Without retraining the SAM backbone, our approach achieves substantial performance gains across three remote sensing benchmarks-LoveDA, iSAID, and WHU-CD-outperforming both zero-shot SAM and task-specific models with minimal computational overhead. Our results highlight the value of domain-adaptive augmentation for foundation models and offer a scalable path toward more robust geospatial segmentation.

Lost in Translation? Vocabulary Alignment for Source-Free Domain Adaptation in Open-Vocabulary Semantic Segmentation

We introduce VocAlign, a novel source-free domain adaptation framework specifically designed for VLMs in open-vocabulary semantic segmentation. Our method adopts a student-teacher paradigm enhanced with a vocabulary alignment strategy, which improves pseudo-label generation by incorporating additional class concepts. To ensure efficiency, we use Low-Rank Adaptation (LoRA) to fine-tune the model, preserving its original capabilities while minimizing computational overhead. In addition, we propose a Top-K class selection mechanism for the student model, which significantly reduces memory requirements while further improving adaptation performance. Our approach achieves a notable 6.11 mIoU improvement on the CityScapes dataset and demonstrates superior performance on zero-shot segmentation benchmarks, setting a new standard for source-free adaptation in the open-vocabulary setting.

FloorSAM: SAM-Guided Floorplan Reconstruction with Semantic-Geometric Fusion

Reconstructing building floor plans from point cloud data is key for indoor navigation, BIM, and precise measurements. Traditional methods like geometric algorithms and Mask R-CNN-based deep learning often face issues with noise, limited generalization, and loss of geometric details. We propose FloorSAM, a framework that integrates point cloud density maps with the Segment Anything Model (SAM) for accurate floor plan reconstruction from LiDAR data. Using grid-based filtering, adaptive resolution projection, and image enhancement, we create robust top-down density maps. FloorSAM uses SAM's zero-shot learning for precise room segmentation, improving reconstruction across diverse layouts. Room masks are generated via adaptive prompt points and multistage filtering, followed by joint mask and point cloud analysis for contour extraction and regularization. This produces accurate floor plans and recovers room topological relationships. Tests on Giblayout and ISPRS datasets show better accuracy, recall, and robustness than traditional methods, especially in noisy and complex settings. Code and materials: github.com/Silentbarber/FloorSAM.

RangeSAM: Leveraging Visual Foundation Models for Range-View repesented LiDAR segmentation

Point cloud segmentation is central to autonomous driving and 3D scene understanding. While voxel- and point-based methods dominate recent research due to their compatibility with deep architectures and ability to capture fine-grained geometry, they often incur high computational cost, irregular memory access, and limited real-time efficiency. In contrast, range-view methods, though relatively underexplored - can leverage mature 2D semantic segmentation techniques for fast and accurate predictions. Motivated by the rapid progress in Visual Foundation Models (VFMs) for captioning, zero-shot recognition, and multimodal tasks, we investigate whether SAM2, the current state-of-the-art VFM for segmentation tasks, can serve as a strong backbone for LiDAR point cloud segmentation in the range view. We present , to our knowledge, the first range-view framework that adapts SAM2 to 3D segmentation, coupling efficient 2D feature extraction with standard projection/back-projection to operate on point clouds. To optimize SAM2 for range-view representations, we implement several architectural modifications to the encoder: (1) a novel module that emphasizes horizontal spatial dependencies inherent in LiDAR range images, (2) a customized configuration of tailored to the geometric properties of spherical projections, and (3) an adapted mechanism in the encoder backbone specifically designed to capture the unique spatial patterns and discontinuities present in range-view pseudo-images. Our approach achieves competitive performance on SemanticKITTI while benefiting from the speed, scalability, and deployment simplicity of 2D-centric pipelines. This work highlights the viability of VFMs as general-purpose backbones for 3D perception and opens a path toward unified, foundation-model-driven LiDAR segmentation. Results lets us conclude that range-view segmentation methods using VFMs leads to promising results.

SCENEFORGE: Enhancing 3D-text alignment with Structured Scene Compositions

The whole is greater than the sum of its parts-even in 3D-text contrastive learning. We introduce SceneForge, a novel framework that enhances contrastive alignment between 3D point clouds and text through structured multi-object scene compositions. SceneForge leverages individual 3D shapes to construct multi-object scenes with explicit spatial relations, pairing them with coherent multi-object descriptions refined by a large language model. By augmenting contrastive training with these structured, compositional samples, SceneForge effectively addresses the scarcity of large-scale 3D-text datasets, significantly enriching data complexity and diversity. We systematically investigate critical design elements, such as the optimal number of objects per scene, the proportion of compositional samples in training batches, and scene construction strategies. Extensive experiments demonstrate that SceneForge delivers substantial performance gains across multiple tasks, including zero-shot classification on ModelNet, ScanObjNN, Objaverse-LVIS, and ScanNet, as well as few-shot part segmentation on ShapeNetPart. SceneForge's compositional augmentations are model-agnostic, consistently improving performance across multiple encoder architectures. Moreover, SceneForge improves 3D visual question answering on ScanQA, generalizes robustly to retrieval scenarios with increasing scene complexity, and showcases spatial reasoning capabilities by adapting spatial configurations to align precisely with textual instructions.