论文检索

Linhua Ye, Xing Xi, Ronghua Luo

Open-world object detection (OWOD) aims to detect known and unknown objects in dynamic environments. However, only known classes are labeled during training, making it challenging for detectors to recognize unknown objects during inference. Existing methods typically rely on supervision from known categories, leading models to overconfidently misclassify visually similar unknowns as known, and dissimilar ones as background. This known-class prior bias limits the model’s ability to detect unknown objects. In this paper, we propose a novel method, OW-DAR, which enhances foreground-background separability through collaborative fine-grained and coarse-grained modeling. At the fine-grained level, we propose Fine-grained Masked Reconstruction (FMR), which randomly masks regions of the feature map to guide the reconstruction toward semantic structures, rather than memorizing low-level patterns. At the coarse-grained level, we propose Adaptive Region-based Error Aggregation (AREA), which operates on object proposals to aggregate reconstruction errors. This enables the model to attend to semantically ambiguous foreground-background boundaries while suppressing the influence of local outliers during optimization. Finally, we leverage robust reconstruction errors to perform unsupervised foreground-background modeling, enabling probabilistic estimation for potential unknown objects. We validate the effectiveness of OW-DAR on standard OWOD benchmark. Experimental results demonstrate that OW-DAR consistently outperforms existing state-of-the-art methods, achieving a +18.8 improvement in unknown object recall (U-Recall).

Kai Ye, YingShi Luan, Zhudi Chen, Guangyue Meng, Pingyang Dai, Liujuan Cao

Referring Image Segmentation (RIS), which aims to segment specific objects based on natural language descriptions, plays an essential role in vision-language understanding. Despite its progress in remote sensing applications, RIS under Low-Altitude Drone (LAD) scenarios remains underexplored, as existing datasets and methods are typically designed for high-altitude and static-view imagery. They struggled to handle the unique characteristics of LAD views, such as diverse viewpoints and high object density. In this paper, we propose RIS-LAD, the first fine-grained RIS benchmark tailored for LAD scenarios, featuring 13,871 meticulously annotated image-text-mask triplets collected from real-world drone footage with emphasis on small, densely cluttered objects and multi-view perspectives. Additionally, we propose the Semantic-Aware Adaptive Reasoning Network, which decomposes and adaptively routes semantic information to different network stages rather than uniformly injecting all linguistic features. Specifically, the Category-Dominated Linguistic Enhancement aligns visual features with object categories during early encoding, while the Adaptive Reasoning Fusion Module dynamically selects semantic cues across scales to enhance reasoning in complex scenes. Extensive experiments reveal that RIS-LAD presents substantial challenges to state-of-the-art RIS algorithms, and also demonstrate the effectiveness of our proposed model in addressing these challenges.

Ziwei Yao, Qian Wang, Ruiping Wang, Xilin Chen

With the advancement of vision-language models, image captioning has made significant progress, leading to the generation of more accurate and detailed descriptions. Current image captioning primarily focuses on describing the apparent visual characteristics, which are easily observed by most humans, but less helpful in real-world scenarios. When users seek a deeper understanding of visual content, they may be concerned with fine-grained categories, function properties, and other background knowledge, rather than merely appearances. Additionally, as users' interests vary, there is a growing demand for customizable content generation. To address these challenges, we propose the task of image narrative generation, which aims to produce knowledge-rich natural language responses for input images, customized to the user preference. Furthermore, we propose T^4, an image narrative generation model progressing through cascade steps: Tailor, reTrieve, Think, and Tell. Specifically, it takes the image and various types of prompts as input, and first refines or predicts potentially interesting queries that are tailored to the user expertise level. Subsequently, the model enriches contextual knowledge through retrieval-augmentation and employs chain-of-thoughts to decompose the generation process step by step, thereby telling an accurate and logically coherent image narrative. In addition, we construct the ImgNarr-23K dataset to support task training and evaluation. Experimental results demonstrate that the proposed approach generates image narratives that better satisfy user requirements, and achieves state-of-the-art performance in knowledge-based VQA tasks without additional finetuning. T^4 presents a promising solution for customized content generation in specialized domains.

Yulin Yao, Kangfeng Zheng, Bin Wu, Chunhua Wu, Jujie Wang, Jiaqi Gao, Minjiao Yang, Dan Luo

Driven by advances in GANs and diffusion models, deepfake content has reached an unprecedented level of photorealism, causing detectors to deteriorate once they leave their training domain. Most prior studies adopt CLIP as the backbone of an image-level binary classifier, yet overlook CLIP’s core strength: text-to-image semantic alignment. Moreover, captions generated by CLIP-CAP lack sufficient high-level semantics to distinguish between authentic and manipulated faces. Deepfake generators often fail to maintain semantic coherence, resulting in contradictions that traditional visual models cannot capture. Existing approaches also intermingle all samples during training and thus lack a systematic, difficulty-aware curriculum. To bridge these gaps, we introduce Semantic- and Frequency-Enhanced (SAFE) deepfake detection, a two-component framework: 1) Semantic-enhanced multimodal alignment. Authenticity cues are injected into CLIP-CAP captions, and low-rank LoRA fine-tuning is applied to CLIP’s visual branch, yielding dual supervision for text–image alignment and forgery discrimination. 2) Dual-score curriculum learning. Fourier Correlation Variance (FCV) measures local spectral consistency and, combined with the loss value, is transformed into a difficulty score that ranks training samples from easy to hard, reducing training time by 23.3% and enhancing generalization. SAFE attains state-of-the-art performance on several cross-dataset and cross-manipulation benchmarks. Ablation studies confirm that semantic enhancement, LoRA fine-tuning, and dual-score curriculum are complementary, jointly delivering substantial gains in open-set generalization.

Wenhao Yao, Zhenxin Li, Shiyi Lan, Zi Wang, Xinglong Sun, Jose M. Alvarez, Zuxuan Wu

Autonomous vehicles must navigate safely in complex driving environments. Imitating a single expert trajectory, as in regression-based approaches, usually does not explicitly assess the safety of the predicted trajectory. Selection-based methods address this by generating and scoring multiple trajectory candidates and predicting the safety score for each. However, they face optimization challenges in precisely selecting the best option from thousands of candidates and distinguishing subtle but safety-critical differences, especially in rare and challenging scenarios. We propose DriveSuprim to overcome these challenges and advance the selection-based paradigm through a coarse-to-fine paradigm for progressive candidate filtering, a rotation-based augmentation method to improve robustness in out-of-distribution scenarios, and a self-distillation framework to stabilize training. DriveSuprim achieves state-of-the-art performance, reaching 93.5% PDMS in NAVSIM v1 and 87.1% EPDMS in NAVSIM v2 without extra data, with 83.02 Driving Score and 60.00 Success Rate on Bench2Drive, demonstrating superior planning capabilities in various driving scenarios.

Wei Yao, Yunlian Sun, Hongwen Zhang, Yebin Liu, Jinhui Tang

Generating high-fidelity full-body human interactions with dynamic objects and static scenes remains a critical challenge in computer graphics and animation. Existing methods for human-object interaction often neglect scene context, leading to implausible penetrations, while human-scene interaction approaches struggle to coordinate fine-grained manipulations with long-range navigation. To address these limitations, we propose HOSIG, a novel framework for synthesizing full-body interactions through hierarchical scene perception. Our method decouples the task into three key components: 1) a scene-aware grasp pose generator that ensures collision-free whole-body postures with precise hand-object contact by integrating local geometry constraints, 2) a heuristic navigation algorithm that autonomously plans obstacle-avoiding paths in complex indoor environments via compressed 2D floor maps and dual-component spatial reasoning, and 3) a scene-guided motion diffusion model that generates trajectory-controlled, full-body motions with finger-level accuracy by incorporating spatial anchors and dual-space gradient-based guidance. Extensive experiments on the TRUMANS dataset demonstrate superior performance over state-of-the-art methods. Notably, our framework supports unlimited motion length through autoregressive generation and requires minimal manual intervention. This work bridges the critical gap between scene-aware navigation and dexterous object manipulation, advancing the frontier of embodied interaction synthesis.

Mingze Yao, Zhiying Jiang, Xianping Fu, Huibing Wang

Underwater Image Enhancement (UIE) focuses on improving visual quality from various underwater scenes. Existing methods simplistically treat various degradations as homogeneous, disregarding their intrinsic connections and causing models to blindly learn, resulting in conflicting optimization goals and visual distortions. To address above limitations, we propose a Conditional Prompt Learning via Degradation Perception (CPLDP) model, which employs conditional prompt as degradation perception priors and guides underwater image enhancement. Specifically, we show that the natural language prompts not only promote distinguishing different degraded images, but also aid in exploring more details with semantic information. Therefore, our method generates five key degradation prompts (green/blue/green-blue color casts, uneven illumination and haze) with conditional prompt learning. Subsequently, considering the intrinsic relationships among different degradations, we employ degradation perceptions as priors and fine-tune the learning strategy to enhance underwater images. During training, an adaptive loss function with multi-degradations is designed, allowing it to effectively handle the task conflicts among multiple underwater degradations. Additionally, we conduct a human visual-based underwater dataset with various degradation types by subjective statistics. Extensive experiments on both full-reference and non-reference datasets demonstrate that our CPLDP can achieve better visual results and outperforms state-of-the-art UIE methods across various degradation scenarios.

Liang Yao, Fan Liu, Hongbo Lu, Chuanyi Zhang, Rui Min, Shengxiang Xu, Shimin Di, Pai Peng

Remote sensing imagery presents vast, inherently unstructured spatial data, necessitating sophisticated reasoning to interpret complex user intents and contextual relationships beyond simple recognition tasks. In this paper, we aim to construct an Earth observation workflow to handle complex queries by reasoning about spatial context and user intent. As a reasoning workflow, it should autonomously explore and construct its own inference paths, rather than being confined to predefined ground‑truth sequences. Ideally, its architecture ought to be unified yet generalized, possessing capabilities to perform diverse reasoning tasks through one model without requiring additional fine-tuning. Existing remote sensing approaches rely on supervised fine-tuning paradigms and task‑specific heads, limiting both autonomous reasoning and unified generalization. To this end, we propose RemoteReasoner, a unified workflow for geospatial reasoning. The design of RemoteReasoner integrates a multi-modal large language model (MLLM) for interpreting user instructions and localizing targets, together with task transformation strategies that enable multi-granularity tasks, including object-, region-, and pixel-level. In contrast to existing methods, our framework is trained with reinforcement learning (RL) to endow the MLLM sufficient reasoning autonomy. At the inference stage, our transformation strategies enable diverse task output formats without requiring task-specific decoders or further fine-tuning. Experiments demonstrated that RemoteReasoner achieves state-of-the-art performance across multi-granularity reasoning tasks. Furthermore, it retains the MLLM's inherent generalization capability, demonstrating robust performance on unseen tasks and categories.

Jianhang Yao, Yongbin Zheng, Siqi Lu, Wanying Xu, Peng Sun

To identify objects beyond predefined categories, open-vocabulary aerial object detection (OVAD) leverages the zero-shot capabilities of visual-language models (VLMs) to generalize from base to novel categories. Existing approaches typically utilize self-learning mechanisms with weak text supervision to generate region-level pseudo-labels to align detectors with VLMs semantic spaces. However, text dependence induces semantic bias, restricting open-vocabulary expansion to text-specified concepts. We propose VK-Det, a visual knowledge-guided open-vocabulary object detection framework without extra supervision. First, we discover and leverage vision encoder's inherent informative region perception to attain fine-grained localization and adaptive distillation. Second, we introduce a novel prototype-aware pseudo-labeling strategy. It models inter-class decision boundaries through feature clustering and maps detection regions to latent categories via prototype matching. This enhances attention to novel objects while compensating for missing supervision. Extensive experiments show state-of-the-art performance, achieving 30.1 mAPᴺ on DIOR and 23.3 mAPᴺ on DOTA, outperforming even extra supervised methods.

Huizai Yao, Sicheng Zhao, Pengteng Li, Yi Cui, Shuo Lu, Weiyu Guo, Yunfan Lu, Yijie Xu, Hui Xiong

Source-Free Object Detection (SFOD) aims to adapt a source-pretrained object detector to a target domain without access to source data. However, existing SFOD methods predominantly rely on internal knowledge from the source model, which limits their capacity to generalize across domains and often results in biased pseudo-labels, thereby hindering both transferability and discriminability. In contrast, Vision Foundation Models (VFMs), pretrained on massive and diverse data, exhibit strong perception capabilities and broad generalization, yet their potential remains largely untapped in the SFOD setting. In this paper, we propose a novel SFOD framework that leverages VFMs as external knowledge sources to jointly enhance feature alignment and label quality. Specifically, we design three VFM-based modules: (1) Patch-weighted Global Feature Alignment (PGFA) distills global features from VFMs using patch-similarity–based weighting to enhance global feature transferability; (2) Prototype-based Instance Feature Alignment (PIFA) performs instance-level contrastive learning guided by momentum-updated VFM prototypes; and (3) Dual-source Enhanced Pseudo-label Fusion (DEPF) fuses predictions from detection VFMs and teacher models via an entropy-aware strategy to yield more reliable supervision. Extensive experiments on six benchmarks demonstrate that our method achieves state-of-the-art SFOD performance, validating the effectiveness of integrating VFMs to simultaneously improve transferability and discriminability.

Haiming Yao, Qiyu Chen, Wei Luo, Zheng Zhang, Jianxing Liao, Wei You

The transfer of knowledge from large-scale pre-trained models to diverse downstream tasks has achieved remarkable success. Beyond the traditional full fine-tuning paradigm, Parameter-Efficient Fine-Tuning (PEFT) has emerged as a more efficient model adaptation approach. However, applying existing PEFT methods to adapt dense vision models, particularly in multi-task settings, remains inadequately explored due to their low efficiency, limited task scalability, and neglect of cross-task fine-tuning interactions. To address these challenges, we propose the Task Dynamic-Synergistic Skill Adaptation, termed TDSS, an efficient and scalable multi-task model adaptation framework for dense visual predictions. TDSS comprises two key components: Task-Dynamic Skill Adapters (TDSA) and Task-Synergistic Adaptation Interaction (TSAI). Specifically, TDSA are inserted in parallel into pre-trained vision models to extract task-specific adapted features through the construction of skill representation experts and task dynamic gating. TSAI is developed to enhance cross-task adaptation interaction by bridging global generic and task-specific adapted features. Extensive experiments on multi-task dense visual predictions demonstrate that TDSS surpasses existing state-of-the-art parameter-efficient fine-tuning methods, while exhibiting remarkable efficiency and scalability in parameters and computational complexity.

Haiming Yao, Wei Luo, Qiyu Chen, Jianxing Liao, Wei You

While adapting pretrained vision models to downstream dense prediction tasks is widely used, current methods often overlook adaptation efficiency, especially in the context of multi-task learning (MTL). Although parameter-efficient fine-tuning (PEFT) methods can enhance parameter efficiency, broader aspects such as GPU memory and training time efficiency remain underexplored. In this paper, we propose a new paradigm that simultaneously achieves efficiency in Parameters, GPU Memory, and Training Time for Multi-Task Dense Vision Adaptation. Specifically, we propose a dual-branch framework, in which a frozen pretrained backbone serves as the generic main branch, and the proposed Bi-Directional Task Adaptation (BDTA) modules are integrated in parallel to form a task bypass branch that extracts adaptation features required by multiple specific tasks. This adaptation module is lightweight, efficient, and does not require backpropagation through the large pre-trained backbone, thus avoiding resource-intensive gradient computations. Moreover, a Mixture of Task Experts mechanism (MoTE) is further proposed to integrate adaptation features across tasks and scales, thereby obtaining more robust representations tailored for dense prediction tasks. On the PASCAL-Context benchmark, our method achieves over 2× relative performance improvement compared to the best prior multi-task PEFT method, while using only ~30% of the parameters, ~50% of the memory, and ~60% of the training time, demonstrating superior overall adaptation efficiency.

Zijiang Yang, Hanqing Chao, Bokai Zhao, Yelin Yang, Yunshuo Zhang, Dongmei Fu, Junping Zhang, Le Lu, Ke Yan, Dakai Jin 等

Nucleus detection and classification (NDC) in histopathology analysis is a fundamental task that underpins a wide range of high-level pathology applications. However, existing methods heavily rely on labor-intensive nucleus-level annotations and struggle to fully exploit large-scale unlabeled data for learning discriminative nucleus representations. In this work, we propose MUSE (MUlti-scale denSE self-distillation), a novel self-supervised learning method tailored for NDC. At its core is NuLo (Nucleus-based Local self-distillation), a coordinate-guided mechanism that enables flexible local self-distillation based on predicted nucleus positions. By removing the need for strict spatial alignment between augmented views, NuLo allows critical cross-scale alignment, thus unlocking the capacity of models for fine-grained nucleus-level representation. To support MUSE, we design a simple yet effective encoder-decoder architecture and a large field-of-view semi-supervised fine-tuning strategy that together maximize the value of unlabeled pathology images. Extensive experiments on three widely used benchmarks demonstrate that MUSE effectively addresses the core challenges of histopathological NDC. The resulting models not only surpass state-of-the-art supervised baselines but also outperform generic pathology foundation models.

Zhiwen Yang, Yuxin Peng

Camera-based 3D semantic scene completion (SSC) plays a crucial role in autonomous driving, enabling voxelized 3D scene understanding for effective scene perception and decision-making. Existing SSC methods have shown efficacy in improving 3D scene representations, but suffer from the inherent input-output dimension gap and annotation-reality density gap, where the 2D planar view from input images with sparse annotated labels leads to inferior prediction of real-world dense occupancy with a 3D stereoscopic view. In light of this, we propose the corresponding High-Dimension High-Density Semantic Scene Completion (HD²-SSC) framework with expanded pixel semantics and refined voxel occupancies. To bridge the dimension gap, a High-dimension Semantic Decoupling module is designed to expand 2D image features along a pseudo third dimension, decoupling coarse pixel semantics from occlusions, and then identify focal regions with fine semantics to enrich image features. To mitigate the density gap, a High-density Occupancy Refinement module is devised with a ``detect-and-refine" architecture to leverage contextual geometric and semantic structures for enhanced semantic density with the completion of missing voxels and correction of erroneous ones. Extensive experiments and analyses on the SemanticKITTI and SSCBench-KITTI-360 datasets validate the effectiveness of our HD²-SSC framework.

Zhichao Yang, Tianjiao Gu, Jianjie Wang, Feiyu Lin, Xiangfei Sheng, Pengfei Chen, Leida Li

The increasing popularity of long Text-to-Image (T2I) generation has created an urgent need for automatic and interpretable models that can evaluate the image-text alignment in long prompt scenarios. However, the existing T2I alignment benchmarks predominantly focus on short prompt scenarios and only provide MOS or Likert scale annotations. This inherent limitation hinders the development of long T2I evaluators, particularly in terms of the interpretability of alignment. In this study, we contribute LongT2IBench, which comprises 14K long text-image pairs accompanied by graph-structured human annotations. Given the detail-intensive nature of long prompts, we first design a Generate-Refine-Qualify annotation protocol to convert them into textual graph structures that encompass entities, attributes, and relations. Through this transformation, fine-grained alignment annotations are achieved based on these granular elements. Finally, the graph-structed annotations are converted into alignment scores and interpretations to facilitate the design of T2I evaluation models. Based on LongT2IBench, we further propose LongT2IExpert, a LongT2I evaluator that enables multi-modal large language models (MLLMs) to provide both quantitative scores and structured interpretations through an instruction-tuning process with Hierarchical Alignment Chain-of-Thought (CoT). Extensive experiments and comparisons demonstrate the superiority of the proposed LongT2IExpert in alignment evaluation and interpretation.

Zhenyu Yang, Gensheng Pei, Tao Chen, Xia Yuan, Haofeng Zhang, Xiangbo Shu, Yazhou Yao

Existing paradigms for remote sensing change detection are caught in a trade-off: CNNs excel at efficiency but lack global context, while Transformers capture long-range dependencies at a prohibitive computational cost. This paper introduces ChangeRWKV, a new architecture that reconciles this conflict. By building upon the Receptance Weighted Key Value (RWKV) framework, our ChangeRWKV uniquely combines the parallelizable training of Transformers with the linear-time inference of RNNs. Our approach core features two key innovations: a hierarchical RWKV encoder that builds multi-resolution feature representation, and a novel Spatial-Temporal Fusion Module (STFM) engineered to resolve spatial misalignments across scales while distilling fine-grained temporal discrepancies. ChangeRWKV not only achieves state-of-the-art performance on the LEVIR-CD benchmark, with an 85.46% IoU and 92.16% F1 score, but does so while drastically reducing parameters and FLOPs compared to previous leading methods. This work demonstrates a new, efficient, and powerful paradigm for operational-scale change detection.

Yutong Yang, Lifu Huang, Yijie Lin, Xi Peng, Mouxing Yang

Vision-Language Models (VLMs) excel at extracting salient visual features from query images, thus exhibiting promising visual recognition performance. However, VLMs would encounter significant degradation in fine-grained scenarios due to their deficiency in distinguishing nuanced differences among candidate categories. As a remedy, we draw inspiration from the ``System 1 & System 2" cognitive theory of humans, paving the way to achieve fine-grained recognition for VLMs. To be specific, we observe that VLMs naturally align with System 1, quickly identifying candidate categories but leaving easily-confused ones unresolved. Based on the observation, we propose System-2 enhanCed visuAl recogNition (SCAN), a novel plug-and-play approach that makes VLMs aware of nuanced differences. In brief, SCAN first specifies and abstracts the discriminative attributes for the confused candidate categories and query images by resorting to off-the-shelf large foundation models, respectively. After that, SCAN adaptively integrates the salient visual features from System 1 with the nuanced differences derived from System 2, resolving confusion in candidates with estimated uncertainty. Extensive experiments on eight widely used fine-grained recognition benchmarks against 10 state-of-the-art baselines verify the effectiveness and superiority of SCAN.

Yue Yang, Song Tang, Qijun Zhao, Hailun Zhang, Xiwen Wang, Zijian Deng

The high cost of synthetic aperture radar (SAR) data acquisition motivates SAR image generation research. However, the data scarcity and SAR's inherent azimuth sensitivity make generative models suffer from severe azimuth overfitting. Most existing methods require supplementary data to work effectively, limiting their practicality. In this paper, we propose SAR-DisentDM, a novel semantic-disentangled diffusion model for limited-data SAR image generation, without requiring any auxiliary resources. We develop a physics-aware diffusion architecture that explicitly models semantic knowledge of SAR images, including intrinsic characteristics, contextual diversity, and measurement randomness. A key innovation is the attention-guided semantic disentanglement (AGSD) module, designed to decouple category-specific features from azimuth-variable scattering patterns. This is achieved by aid of a dual disentangled loss with time-step-adaptive optimization. Furthermore, we introduce an azimuth angle perturbation augmentation (AAPA) mechanism, to enhance the model's robustness to minor azimuth angle errors. Extensive evaluations validate that SAR-DisentDM enables controllable SAR image synthesis with designated attributes, significantly improving representation and generalization abilities under limited data. Synthetic imagery from our approach boosts automatic target recognition (ATR) accuracy beyond state-of-the-art methods.

Yitong Yang, Yinglin Wang, Changshuo Wang, Huajie Wang, Shuting He

The success of 3DGS in generative and editing applications has sparked growing interest in 3DGS-based style transfer. However, current methods still face two major challenges: (1) multi-view inconsistency often leads to style conflicts, resulting in appearance smoothing and distortion; and (2) heavy reliance on VGG features, which struggle to disentangle style and content from style images, often causing content leakage and excessive stylization. To tackle these issues, we introduce FantasyStyle, a 3DGS-based style transfer framework, and the first to rely entirely on diffusion model distillation. It comprises two key components: (1) Multi-View Frequency Consistency. We enhance cross-view consistency by applying a 3D filter to multi-view noisy latent, selectively reducing low-frequency components to mitigate stylized prior conflicts. (2) Controllable Stylized Distillation. To suppress content leakage from style images, we introduce negative guidance to exclude undesired content. In addition, we identify the limitations of Score Distillation Sampling and Delta Denoising Score in 3D style transfer and remove the reconstruction term accordingly. Building on these insights, we propose a controllable stylized distillation that leverages negative guidance to more effectively optimize the 3D Gaussians. Extensive experiments demonstrate that our method consistently outperforms state-of-the-art approaches, achieving higher stylization quality and visual realism across various scenes and styles.

Ying Yang, Hui Yin, Aixin Chong, Hui Wang, Zhengyin Liang

Multimodal 3D object detection for autonomous driving, a task for real-world applications, poses substantial challenges in maintaining robust performance under various perturbations and complex environmental conditions. However, most existing approaches primarily focus on performance optimization under relatively ideal scenarios or focus on one or few disturbing conditions (or adverse conditions), lacking systematic exploration of robustness against real-world factors, including high class imbalance, adverse weather conditions, sensor jitter and failures, and significant scene variations. To address this issue, we propose a robust multimodal 3D detector, termed RobusTor3D, which integrates robustness at both the structural and supervisory levels by blending the knowledge from Vision-Language Models (VLMs). Structurally, textual descriptions are incorporated to enhance the semantic richness and diversity of rare classes. This novel semantic injection operation compensates for the inherent class imbalance and modality weakness in conventional visual features. Furthermore, semantic alignment capability and robust representation by Vision-Language Knowledge Extraction (V-LKE) serve as semantic priors to complement modality-specific representations, significantly improving model adaptability. At the supervisory level, we propose a Scene-level Multimodal Consistency Learning (SMCL) strategy, which jointly enforces global semantic constraints across modalities, encouraging the learning of stable and abundant semantic representations. This special design specifically reduces the impact of spatial alignment, while notably enabling semantic compensation under modality-loss conditions. Extensive robustness experiments conducted on KITTI, KITTI-C, and CADC benchmarks evaluate five robustness aspects, including long-tail problem, adverse weather (rain, snow, fog, strong sunlight), sensor spatial misalignment and motion blur, modality loss, and cross-domain scenarios. The results show that RobusTor3D demonstrates superior robustness across all five evaluated aspects. It consistently outperforms the state-of-the-art methods under various challenging conditions.