视觉-语言-动作（VLA）

This paper introduces ZeroMatch, a novel zero-shot RGB-D point cloud registration framework, aimed at achieving robust 3D matching on unseen data without any task-specific training. Our core idea is to utilize the powerful zero-shot image representation of Stable Diffusion, achieved through extensive pre-training on large-scale data, to enhance point-cloud geometric descriptors for robust matching. Specifically, we combine the handcrafted geometric descriptor FPFH with Stable-Diffusion features to create point descriptors that are both locally and contextually aware, enabling reliable RGB-D registration with zero-shot capability. This approach is based on our observation that Stable-Diffusion features effectively encode discriminative global contextual cues, naturally alleviating the feature ambiguity that FPFH often encounters in scenes with repetitive patterns or low overlap. To further enhance cross-view consistency of Stable-Diffusion features for improved matching, we propose a coupled-image input mode that concatenates the source and target images into a single input, replacing the original single-image mode. This design achieves both inter-image and prompt-to-image consistency attentions, facilitating robust cross-view feature interaction and alignment. Finally, we leverage feature nearest neighbors to construct putative correspondences for hypothesize-and-verify transformation estimation. Extensive experiments on 3DMatch, ScanNet, and ScanLoNet verify the excellent zero-shot matching ability of our method.

Visual grounding seeks to localize the image region corresponding to a free-form text description. Recently, the strong multimodal capabilities of Large Vision-Language Models (LVLMs) have driven substantial improvements in visual grounding, though they inevitably require fine-tuning and additional model components to explicitly generate bounding boxes or segmentation masks. However, we discover that a few attention heads in frozen LVLMs demonstrate strong visual grounding capabilities. We refer to these heads, which consistently capture object locations related to text semantics, as localization heads. Using localization heads, we introduce a straightforward and effective training-free visual grounding framework that utilizes text-to-image attention maps from localization heads to identify the target objects. Surprisingly, only three out of thousands of attention heads are sufficient to achieve competitive localization performance compared to existing LVLM-based visual grounding methods that require fine-tuning. Our findings suggest that LVLMs can innately ground objects based on a deep comprehension of the text-image relationship, as they implicitly focus on relevant image regions to generate informative text outputs.

Vision-Language Models (VLMs) excel in integrating visual and textual information for vision-centric tasks, but their handling of inconsistencies between modalities is underexplored. We investigate VLMs' modality preferences when faced with visual data and varied textual inputs in vision-centered settings.By introducing textual variations to four vision-centric tasks and evaluating ten Vision-Language Models (VLMs), we discover a "blind faith in text" phenomenon: VLMs disproportionately trust textual data over visual data when inconsistencies arise, leading to significant performance drops under corrupted text and raising safety concerns.We analyze factors influencing this text bias, including instruction prompts, language model size, text relevance, token order, and the interplay between visual and textual certainty. While certain factors, such as scaling up the language model size, slightly mitigate text bias, others like token order can exacerbate it due to positional biases inherited from language models. To address this issue, we explore supervised fine-tuning with text augmentation and demonstrate its effectiveness in reducing text bias. Additionally, we provide a theoretical analysis suggesting that the blind faith in text phenomenon may stem from an imbalance of pure text and multi-modal data during training.Our findings highlight the need for balanced training and careful consideration of modality interactions in VLMs to enhance their robustness and reliability in handling multi-modal data inconsistencies.

Vision-Language Models (VLMs) have recently demonstrated remarkable capabilities in comprehending complex visual content. However, the mechanisms underlying how VLMs process visual information remain largely unexplored. In this paper, we conduct a thorough empirical analysis, focusing on the attention modules across layers, by which we reveal several key insights about how these models process visual data: (i) the internal representation of the query tokens (e.g., representations of "describe the image"), is utilized by the model to store global image information; we demonstrate that the model generates surprisingly descriptive responses solely from these tokens, without direct access to image tokens. (ii) Cross-modal information flow is predominantly influenced by the middle layers (approximately 25% of all layers), while early and late layers contribute only marginally. (iii) Fine-grained visual attributes and object details are directly extracted from image tokens in a spatially localized manner, i.e., the generated tokens associated with a specific object or attribute attend strongly to their corresponding regions in the image. We propose novel quantitative evaluation to validate our observations, leveraging real-world complex visual scenes. Finally, we demonstrate the potential of our findings in facilitating efficient visual processing in state-of-the-art VLMs.

Recent breakthroughs in Multimodal Large Language Models (MLLMs) have gained significant recognition within the deep learning community, where the fusion of the Video Foundation Models (VFMs) and Large Language Models(LLMs) has proven instrumental in constructing robust video understanding systems, effectively surmounting constraints associated with predefined visual tasks. These sophisticated MLLMs exhibit remarkable proficiency in comprehending videos, swiftly attaining unprecedented performance levels across diverse benchmarks. However, their operation demands substantial memory and computational resources, underscoring the continued importance of traditional models in video comprehension tasks. In this paper, we introduce a novel learning paradigm termed MLLM4WTAL. This paradigm harnesses the potential of MLLM to offer temporal action key semantics and complete semantic textual cues for conventional Weakly-supervised Temporal Action Localization (WTAL) methods. MLLM4WTAL facilitates the enhancement of WTAL by leveraging MLLM guidance. It achieves this by integrating two distinct modules: Key Semantic Matching (KSM) and Complete Semantic Reconstruction (CSR). These modules work in tandem to effectively address prevalent issues like incomplete and over-complete outcomes common in WTAL methods. Rigorous experiments are conducted to validate the efficacy of our proposed approach in augmenting the performance of various heterogeneous WTAL models.

Recent advancements in vision-language models have enhanced performance by increasing the length of visual tokens, making them much longer than text tokens and significantly raising computational costs.However, we observe that the visual tokens generated by popular vision encoders, such as CLIP and SigLIP, contain significant redundancy. To address this, we introduce VisionZip, a simple yet effective method that selects a set of informative tokens for input to the language model, reducing visual token redundancy and improving efficiency while maintaining model performance. The proposed VisionZip can be widely applied to image and video understanding tasks and is well-suited for multi-turn dialogues in real-world scenarios, where previous methods tend to underperform.Experimental results show that VisionZip outperforms the previous state-of-the-art method by at least 5% performance gains across nearly all settings.Moreover, our method significantly enhances model inference speed, improving the prefilling time by 8xand enabling the LLaVA-Next 13B model to infer faster than the LLaVA-Next 7B model while achieving better results.Furthermore, we analyze the causes of this redundancy and encourage the community to focus on extracting better visual features rather than merely increasing token length. Our code is available at https://github.com/dvlab-research/VisionZip.

This paper introduces VisionPAD, a novel self-supervised pre-training paradigm designed for vision-centric algorithms in autonomous driving. In contrast to previous approaches that employ neural rendering with explicit depth supervision, VisionPAD utilizes more efficient 3D Gaussian Splatting to reconstruct multi-view representations using only images as supervision. Specifically, we introduce a self-supervised method for voxel velocity estimation. By warping voxels to adjacent frames and supervising the rendered outputs, the model effectively learns motion cues in the sequential data. Furthermore, we adopt a multi-frame photometric consistency approach to enhance geometric perception. It projects adjacent frames to the current frame based on rendered depths and relative poses, boosting the 3D geometric representation through pure image supervision. Extensive experiments on autonomous driving datasets demonstrate that VisionPAD significantly improves performance in 3D object detection, occupancy prediction and map segmentation, surpassing state-of-the-art pre-training strategies by a considerable margin.

Many practical vision-language applications require models that understand negation, e.g., when using natural language to retrieve images which contain certain objects but not others. Despite advancements in vision-language models (VLMs) through large-scale training, their ability to comprehend negation remains underexplored. This study addresses the question: how well do current VLMs understand negation? We introduce \benchmark , a new benchmark designed to evaluate negation understanding across 18 task variations and 79k examples spanning image, video, and medical datasets. The benchmark consists of two core tasks designed to evaluate negation understanding in diverse multimodal settings: Retrieval with Negation and Multiple Choice Questions with Negated Captions. Our evaluation reveals that modern VLMs struggle significantly with negation, often performing at chance level. To address these shortcomings, we explore a data-centric approach wherein we finetune CLIP models on large-scale synthetic datasets containing millions of negated captions. We show that this approach can result in a 10% increase in recall on negated queries and a 28% boost in accuracy on multiple-choice questions with negated captions.

Vision-language models (VLMs) like CLIP (Contrastive Language-Image Pre-Training) have seen remarkable success in visual recognition, highlighting the increasing need to safeguard the intellectual property (IP) of well-trained models. Effective IP protection extends beyond ensuring authorized usage; it also necessitates restricting model deployment to authorized data domains, particularly when the model is fine-tuned for specific target domains. However, current IP protection methods often rely solely on the visual backbone, which may lack sufficient semantic richness. To bridge this gap, we introduce IP-CLIP, a lightweight IP protection strategy tailored to CLIP, employing a prompt-based learning approach. By leveraging the frozen visual backbone of CLIP, we extract both image style and content information, incorporating them into the learning of IP prompt. This strategy acts as a robust barrier, effectively preventing the unauthorized transfer of features from authorized domains to unauthorized ones. Additionally, we propose a style-enhancement branch that constructs feature banks for both authorized and unauthorized domains. This branch integrates self-enhanced and cross-domain features, further strengthening IP-CLIP's capability to block features from unauthorized domains. Finally, we present new three metrics designed to better balance the performance degradation of authorized and unauthorized domains. Comprehensive experiments in various scenarios demonstrate its promising potential for application in IP protection tasks for VLMs.

Dynamic image degradations, including noise, blur and lighting inconsistencies, pose significant challenges in image restoration, often due to sensor limitations or adverse environmental conditions. Existing Deep Unfolding Networks (DUNs) offer stable restoration performance but require manual selection of degradation matrices for each degradation type, limiting their adaptability across diverse scenarios.To address this issue, we propose the Vision-Language-guided Unfolding Network (VLU-Net), a unified DUN framework for handling multiple degradation types simultaneously.VLU-Net leverages a Vision-Language Model (VLM) refined on degraded image-text pairs to align image features with degradation descriptions, selecting the appropriate transform for target degradation.By integrating an automatic VLM-based gradient estimation strategy into the Proximal Gradient Descent (PGD) algorithm, VLU-Net effectively tackles complex multi-degradation restoration tasks while maintaining interpretability. Furthermore, we design a hierarchical feature unfolding structure to enhance VLU-Net framework, efficiently synthesizing degradation patterns across various levels.VLU-Net is the first all-in-one DUN framework and outperforms current leading one-by-one and all-in-one end-to-end methods by 3.74 dB on the SOTS dehazing dataset and 1.70 dB on the Rain100L deraining dataset.

Depth estimation is a core problem in robotic perception and vision tasks, but 3D reconstruction from a single image presents inherent uncertainties. Current depth estimation models primarily rely on inter-image relationships for supervised training, often overlooking the intrinsic information provided by the camera itself. We propose a method that embodies the camera model and its physical characteristics into a deep learning model, computing embodied scene depth through real-time interactions with road environments. The model can calculate embodied scene depth in real-time based on immediate environmental changes using only the intrinsic properties of the camera, without any additional equipment. By combining embodied scene depth with RGB image features, the model gains a comprehensive perspective on both geometric and visual details. Additionally, we incorporate text descriptions containing environmental content and depth information as priors for scene understanding, enriching the model's perception of objects. This integration of image and language -- two inherently ambiguous modalities -- leverages their complementary strengths for monocular depth estimation. The real-time nature of the embodied language and depth prior model ensures that the model can continuously adjust its perception and behavior in dynamic environments. Experimental results show that the embodied depth estimation method enhances model performance across different scenes.

Recent advances in human motion prediction (HMP) have shifted focus from isolated motion data to integrating human-scene correlations. In particular, the latest methods leverage human gaze points, using their spatial coordinates to indicate intent--where a person might move within a 3D environment. Despite promising trajectory results, these methods often produce inaccurate poses by overlooking the semantic implications of gaze, specifically the affordances of observed objects, which indicate the possible interactions. To address this, we propose GAP3DS, an affordance-aware HMP model that utilizes gaze-informed object affordances to improve HMP in complex 3D environments. GAP3DS incorporates a gaze-guided affordance learner to identify relevant objects in the scene and infer their affordances based on human gaze, thus contextualizing future human-object interactions. This affordance information, enriched with visual features and gaze data, conditions the generation of multiple human-object interaction poses, which are subsequently decoded into final motion predictions. Extensive experiments on two real-world datasets demonstrate that GAP3DS outperforms state-of-the-art methods in both trajectory and pose accuracy, producing more physically consistent and contextually grounded predictions. For more details and code, please refer to the project page.

Vision-language foundation models have shown impressive capabilities across various zero-shot tasks, including training-free localization and grounding, primarily focusing on localizing objects in images. However, leveraging those capabilities to localize actions and events in videos is challenging, as actions have less physical outline and are usually described by higher-level concepts.In this work, we propose VideoGEM, the first training-free spatial action grounding method based on pretrained image- and video-language backbones. Namely, we adapt the self-self attention formulation of GEM to spatial activity grounding. We observe that high-level semantic concepts, such as actions, usually emerge in the higher layers of the image- and video-language models. We, therefore, propose a layer weighting in the self-attention path to prioritize higher layers. Additionally, we introduce a dynamic weighting method to automatically tune layer weights to capture each layer's relevance to a specific prompt. Finally, we introduce a prompt decomposition, processing action, verb, and object prompts separately, resulting in a better spatial localization of actions. We evaluate the proposed approach on three image- and video-language backbones, CLIP, OpenCLIP, and ViCLIP, and on four video grounding datasets, V-HICO, DALY, YouCook-Interactions, and GroundingYouTube, showing that the proposed training-free approach is able to outperform current trained state-of-the-art approaches for spatial video grounding.

The recent surge in high-quality visual instruction tuning samples from closed-source vision-language models (VLMs) such as GPT-4V has accelerated the release of open-source VLMs across various model sizes. However, scaling VLMs to improve performance using larger models brings significant computational challenges, especially for deployment on resource-constrained devices like mobile platforms and robots. To address this, we propose VLsI: Verbalized Layers-to-Interactions, a new VLM family in 2B and 7B model sizes, which prioritizes efficiency without compromising accuracy. VLsI leverages a unique, layer-wise distillation process, introducing intermediate "verbalizers" that map features from each layer to natural language space, allowing smaller VLMs to flexibly align with the reasoning processes of larger VLMs. This approach mitigates the training instability often encountered in output imitation and goes beyond typical final-layer tuning by aligning the small VLMs' layer-wise progression with that of the large ones. We validate VLsI across ten challenging vision-language benchmarks, achieving notable performance gains (11.0% for 2B and 17.4% for 7B) over GPT-4V without the need for model scaling, merging, or architectural changes.

Vision-based Reinforcement Learning (VRL) attempts to establish associations between visual inputs and optimal actions through interactions with the environment. Given the high-dimensional and complex nature of visual data, it becomes essential to learn policy upon high-quality state representation. To this end, existing VRL methods primarily rely on interaction-collected data, combined with self-supervised auxiliary tasks. However, two key challenges remain: limited data samples and a lack of task-relevant semantic constraints. To tackle this, we propose DGC, a method that distills guidance from Visual Language Models (VLMs) alongside self-supervised learning into a compact VRL agent. Notably, we leverage the state representation capabilities of VLMs, rather than their decision-making abilities. Within DGC, a novel prompting-reasoning pipeline is designed to convert historical observations and actions into usable supervision signals, enabling semantic understanding within the compact visual encoder. By leveraging these distilled semantic representations, the VRL agent achieves significant improvements in the sample efficiency. Extensive experiments on the Carla benchmark demonstrate our state-of-the-art performance.

Vision-language generative reward models (VL-GenRMs) play a crucial role in aligning and evaluating multimodal AI systems, yet their own evaluation remains under-explored. Current assessment methods primarily rely on AI-annotated preference labels from traditional VL tasks, which can introduce biases and often fail to effectively challenge state-of-the-art models. To address these limitations, we introduce VL-RewardBench, a comprehensive benchmark spanning general multimodal queries, visual hallucination detection, and complex reasoning tasks. Through our AI-assisted annotation pipeline that combines sample selection with human verification, we curate 1,250 high-quality examples specifically designed to probe VL-GenRMs limitations. Comprehensive evaluation across 16 leading large vision-language models demonstrates VL-RewardBench's effectiveness as a challenging testbed, where even GPT-4o achieves only 65.4% accuracy, and state-of-the-art open-source models such as Qwen2-VL-72B, struggle to surpass random-guessing. Importantly, performance on VL-RewardBench strongly correlates (Pearson's r > 0.9) with MMMU-Pro accuracy using Best-of-N sampling with VL-GenRMs. Analysis experiments uncover three critical insights for improving VL-GenRMs: (i) models predominantly fail at basic visual perception tasks rather than reasoning tasks; (ii) inference-time scaling benefits vary dramatically by model capacity; and (iii) training VL-GenRMs to learn to judge substantially boosts judgment capability (+14.7% accuracy for a 7B VL-GenRM). We believe VL-RewardBench along with the experimental insights will become a valuable resource for advancing VL-GenRMs. Project page: https://vl-rewardbench.github.io

Generalist vision language models (VLMs) have made significant strides in computer vision, but they fall short in specialized fields like healthcare, where expert knowledge is essential. Current large multimodal models like Gemini and GPT-4o are insufficient for medical tasks due to their reliance on memorized internet knowledge rather than the nuanced expertise required in healthcare. Meanwhile, existing medical VLMs (e.g. Med-Gemini) often lack expert consultation as part of their design, and many rely on outdated, static datasets that were not created with modern, large deep learning models in mind. VLMs are usually trained in three stages: vision pre-training, vision-language pre-training, and instruction fine-tuning (IFT). IFT has been typically applied using a mixture of generic and healthcare data. In contrast, we propose that for medical VLMs, a fourth stage of specialized IFT is necessary, which focuses on medical data and includes information from domain expert models. Domain expert models developed for medical use are crucial because they are specifically trained for certain clinical tasks, e.g. to detect tumors and classify abnormalities through segmentation and classification, which learn fine-grained features of medical data-features that are often too intricate for a VLM to capture effectively. This paper introduces a new framework, VILA-M3, for medical VLMs that utilizes domain knowledge via expert models. We argue that generic VLM architectures alone are not viable for real-world clinical applications and on-demand usage of domain-specialized expert model knowledge is critical for advancing AI in healthcare. Through our experiments, we show an improved state-of-the-art (SOTA) performance with an average improvement of ~9% over the prior SOTA model Med-Gemini and ~6% over models trained on the specific tasks. Our approach emphasizes the importance of domain expertise in creating precise, reliable VLMs for medical applications.

Vision-Language Models (VLMs) demand substantial computational resources during inference, largely due to the extensive visual input tokens for representing visual information. Previous studies have noted that visual tokens tend to receive less attention than text tokens, suggesting their lower importance during inference and potential for pruning. However, their methods encounter several chal- lenges: reliance on greedy heuristic criteria for token im- portance and incompatibility with FlashAttention and KV cache. To address these issues, we introduce TopV, a com- patible TOken Pruning with inference Time Optimization for fast and low-memory VLM, achieving efficient pruning without additional training or fine-tuning. Instead of re- lying on attention scores, we formulate token pruning as an optimization problem, accurately identifying important visual tokens while remaining compatible with FlashAtten- tion. Additionally, since we only perform this pruning once during the prefilling stage, it effectively reduces KV cache size. Our optimization framework incorporates a visual- aware cost function considering factors such as Feature Similarity, Relative Spatial Distance, and Absolute Central Distance, to measure the importance of each source visual token, enabling effective pruning of low-importance tokens. Extensive experiments demonstrate that our method outperforms previous token pruning methods, validating the effectiveness and efficiency of our approach.

Temporal action detection (TAD) aims to identify and localize action instances in untrimmed videos, which is essential for various video understanding tasks. However, recent improvements in model performance, driven by larger feature extractors and datasets, have led to increased computational demands. This presents a challenge for applications like autonomous driving and robotics, which rely on limited computational resources. While existing channel pruning methods can compress these models, reducing the number of channels often hinders the parallelization efficiency of GPU, due to the inefficient multiplication between small matrices. Instead of pruning channels, we propose a **Progressive Block Drop** method that reduces model depth while retaining layer width. In this way, we still use large matrices for computation but reduce the number of multiplications. Our approach iteratively removes redundant blocks in two steps: first, we drop blocks with minimal impact on model performance; and second, we employ a parameter-efficient cross-depth alignment technique, fine-tuning the pruned model to restore model accuracy. Our method achieves a 25% reduction in computational overhead on two TAD benchmarks (THUMOS14 and ActivityNet-1.3) to achieve lossless compression. More critically, we empirically show that our method is orthogonal to channel pruning methods and can be combined with it to yield further efficiency gains.

When a vision-language model (VLM) is prompted to identify an entity depicted in an image, it may answer "I see a conifer," rather than the specific label "Norway spruce". This raises two issues for evaluation: Firstly, the unconstrained generated text needs to be mapped to the evaluation label space (i.e., "conifer"). Secondly, a useful classification measure should give partial credit to less specific, but not incorrect, answers ("Norway spruce" being a type of "conifer"). To meet these requirements, we propose a framework for evaluating unconstrained text predictions such as those generated from a vision-language model against a taxonomy. Specifically, we propose the use of hierarchical precision and recall measures to assess the level of correctness and specificity of predictions with regard to a taxonomy. Experimentally, we first show that existing text similarity measures do not capture taxonomic similarity well. We then develop and compare different methods to map textual VLM predictions onto a taxonomy. This allows us to compute hierarchical similarity measures between the generated text and the ground truth labels. Finally, we analyze modern VLMs on fine-grained visual classification tasks based on our proposed taxonomic evaluation scheme.