Generative Zero-Shot Learning (GZSL) methods address the challenge of recognizing unseen classes by synthesizing visual features, thereby converting ZSL into a supervised learning task. However, existing approaches are predominantly constrained to two multi-stage strategies: pre-generation prior knowledge enhancement and post-generation feature refinement. These paradigms often suffer from error propagation across stages, ultimately limiting generation quality and representational fidelity. To overcome these limitations, we propose DKGZSL, a novel generative framework that injects dynamic visual-semantic knowledge directly into the feature synthesis process, effectively unifying generation and refinement into a single cohesive stage. Specifically, a Knowledge Transfer Network (KTN) is introduced to convert semantic information into hierarchical visual knowledge representations. To ensure accurate semanticvisual alignment, we further design a Semantic-Oriented Visual Refinement (SOVR) module that reshapes real visual features into semantically aligned and noise-suppressed representations, providing precise guidance for the KTN. Moreover, hierarchical knowledge extracted from each KTN layer is progressively transmitted to the generator via Meta-Fusion Units (MFUs), enabling dynamic semantic guidance and improving generation quality. Extensive experiments on three benchmark datasets demonstrate that DKGZSL achieves consistent state-of-the-art performance with both ResNet-101 and ViT-B/16 feature extractors. Comprehensive ablation studies further confirm the effectiveness and complementarity of each proposed component. The code is available at https://github.com/JingHu-gdut/DKGZSL.

This paper argues that Large Language Models (LLMs) should incorporate explicit mechanisms for human empathy. As LLMs become increasingly deployed in high-stakes human-centered settings, their success depends not only on correctness or fluency but on faithful preservation of human perspectives. Yet, current LLMs systematically fail at this requirement: even when well-aligned and policy-compliant, they often attenuate affect, misrepresent contextual salience, and rigidify relational stance in ways that distort meaning. We formalize empathy as an observable behavioral property: the capacity to model and respond to human perspectives while preserving intention, affect, and context. Under this framing, we identify four recurring mechanisms of empathic failure in contemporary LLMs–sentiment attenuation, empathic granularity mismatch, conflict avoidance, and linguistic distancing–arising as structural consequences of prevailing training and alignment practices. We further organize these failures along three dimensions: cognitive, cultural, and relational empathy, to explain their manifestation across tasks. Empirical analyses show that strong benchmark performance can mask systematic empathic distortions, motivating empathy-aware objectives, benchmarks, and training signals as first-class components of LLM development.

The explosive growth of multimedia data has established cross-modal retrieval as a critical research frontier, with cross-modal hashing (CMH) emerging as an effective technique for managing large-scale datasets. However, existing CMH methods typically rely on the unrealistic assumption that all multimodal training data are complete. This paper addresses the fundamental challenge of insufficient correlation information in incomplete multimodal scenarios, which manifests as two interconnected problems, i.e., intra-class dispersion caused by inadequate intra-modal correlation and limited inter-class separability resulting from insufficient cross-modal correlation. To overcome these limitations, we propose Semantic Consistent Memory Hashing with Incremental Regularization (SCMIR), a novel framework for incomplete cross-modal retrieval. Specifically, we first establish a robust semantic-consistent hash learning framework based on information bottleneck theory for complete paired samples, learning compact representations that preserve essential category information while eliminating modality-specific redundancy. We then introduce a memory bank-based incremental regularization mechanism that stores high-quality hash codes from complete data as semantic references, enabling effective knowledge transfer to incomplete samples through alignment constraints. Extensive experiments on three benchmark datasets demonstrate that SCMIR significantly outperforms state-of-the-art methods across various hash code lengths, achieving superior performance in both text-to-image and image-to-text retrieval tasks.

While 3D Gaussian Splatting (3DGS) has demonstrated impressive real-time rendering performance, its efficacy remains constrained by a reliance on heuristic density control. Despite numerous refinements to these handcrafted rules, such methods inherently lack the flexibility to adapt to diverse scenes with complex geometries. In this paper, we propose a paradigm shift for density control from rigid heuristics to fully learnable policies. Specifically, we introduce LeGS, a framework that reformulates density control as a parameterized policy network optimized via Reinforcement Learning (RL). Central to our approach is the tailored effective reward function grounded in sensitivity analysis, which precisely quantifies the marginal contribution of individual Gaussians to reconstruction quality. To maintain computational tractability, we derive a closed-form solution that reduces the complexity of reward calculation from O(N^2) to O(N). Extensive experiments on the Mip-NeRF 360, Tanks & Temples, and Deep Blending datasets demonstrate that LeGS significantly outperforms state-of-the-art methods, striking a superior balance between reconstruction quality and efficiency. The code will be released at https://github.com/AaronNZH/LeGS

Deep research agents powered by Large Language Models (LLMs) can perform multi-step reasoning, web exploration, and long-form report generation. However, existing systems remain largely autonomous, assuming fully specified user intent and evaluating only final outputs. In practice, research goals are often underspecified and evolve during exploration, yet current benchmarks neither model dynamic user feedback nor measure interaction costs. To address this gap, we introduce IDRBench, the first Interactive Deep Research Benchmark for systematically evaluating the interactive capabilities of deep research agents. IDRBench formulates deep research as an interactive process where agents may solicit clarification to better align with user intent. It integrates a modular interactive framework, a scalable reference-grounded user simulator, and an interaction-aware evaluation suite that jointly measures alignment gains and interaction overhead. Experiments on seven representative proprietary and open-weight LLMs show that interaction consistently improves research quality and robustness, while revealing substantial differences in interaction efficiency across models. These findings establish interactive capability as a distinct evaluation dimension and position IDRBench as a reusable benchmark for future user-aligned deep research agents.

Dynamic Gaussian Splatting approaches have achieved remarkable performance for 4D scene reconstruction. However, these approaches rely on dense-frame video sequences for photorealistic reconstruction. In real-world scenarios, due to equipment constraints, sometimes only sparse frames are accessible. In this paper, we propose Sparse4DGS, the first method for sparse-frame dynamic scene reconstruction. We observe that dynamic reconstruction methods fail in both canonical and deformed spaces under sparse-frame settings, especially in areas with high texture richness. Sparse4DGS tackles this challenge by focusing on texture-rich areas. For the deformation network, we propose Texture-Aware Deformation Regularization, which introduces a texture-based depth alignment loss to regulate Gaussian deformation. For the canonical Gaussian field, we introduce Texture-Aware Canonical Optimization, which incorporates texture-based noise into the gradient descent process of canonical Gaussians. Extensive experiments show that when taking sparse frames as inputs, our method outperforms existing dynamic or few-shot techniques on NeRF-Synthetic, HyperNeRF, NeRF-DS, and our iPhone-4D datasets.

Large language models accumulate extensive parametric knowledge through pre-training. However, knowledge conflicts occur when outdated or incorrect parametric knowledge conflicts with external knowledge in the context. Existing methods address knowledge conflicts through contrastive decoding, but in conflict-free scenarios, static approaches disrupt output distribution. Other dynamic decoding methods attempt to measure the degree of conflict but still struggle with complex real-world situations. In this paper, we propose a two-stage decoding method called Dynamic Cognitive Reconciliation Decoding (DCRD), to predict and mitigate context-memory conflicts. DCRD first analyzes the attention map to assess context fidelity and predict potential conflicts. Based on this prediction, the input is directed to one of two decoding paths: (1) greedy decoding, or (2) context fidelity-based dynamic decoding. This design enables DCRD to handle conflicts efficiently while maintaining high accuracy and decoding efficiency in conflict-free cases. Additionally, to simulate scenarios with frequent knowledge updates, we constructed ConflictKG, a knowledge conflict QA benchmark. Experiments on four LLMs across six QA datasets show that DCRD outperforms all baselines, achieving state-of-the-art performance.

The deployment of efficient long-context LLMs in applications like autonomous agents, long-chain reasoning, and creative writing is fundamentally bottlenecked by the linear growth of KV cache memory. Existing compression and eviction methods often struggle to balance accuracy, compression ratio, and hardware efficiency. We propose DeltaKV, a residual-based KV cache compression framework motivated by two empirical findings: long-range inter-token similarity and highly shared latent components in KV representations. Instead of discarding tokens, DeltaKV encodes semantic residuals relative to retrieved historical references, preserving fidelity while substantially reducing storage. To translate compression gains into real system speedups, we further introduce Sparse-vLLM, a high-performance inference engine with decoupled memory management and kernels optimized for sparse and irregular KV layouts. Experiments show that DeltaKV reduces KV cache memory to 29\% of the original while maintaining near-lossless accuracy on LongBench, SCBench, and AIME. When integrated with Sparse-vLLM, it achieves up to 2$\times$ throughput improvement over vLLM in long-context scenarios, demonstrating a practical path toward scalable long-context LLM deployment. Code, model checkpoints, and datasets are available at https://github.com/CURRENTF/Sparse-vLLM.

Chest X-ray (CXR) plays a pivotal role in clinical diagnosis, and a variety of task-specific and foundation models have been developed for CXR interpretation. However, these models often struggle to adapt to new diagnostic tasks and complex reasoning scenarios. Recently, LLM-based agents have emerged as a promising paradigm for CXR analysis, enhancing model’s capability via tool coordination, multi-step reasoning, and team collaboration, etc. However, existing agents often rely on a single diagnostic pipeline and lack mechanisms for assessing tools’ reliability, limiting their adaptability and credibility. To this end, we propose CXRAgent, a director-orchestrated, multi-stage agent for CXR interpretation, where a central director coordinates the following stages: (1) Tool Invocation: The agent strategically orchestrates a set of CXR-analysis tools, with outputs normalized and verified by the Evidence-driven Validator (EDV), grounding diagnostic outputs with visual evidence to support reliable downstream diagnosis; (2) Diagnostic Planning: Guided by task requirements and intermediate findings, the agent formulates a targeted diagnostic plan, assembles an expert team, defines member roles, and coordinates their interactions to enable adaptive collaborative reasoning; (3) Collaborative Decision-making: The agent integrates insights from the expert team with accumulated contextual memories, synthesizing them into an evidence-backed conclusion. Experiments on diverse tasks show that CXRAgent achieves strong performance with reliable visual grounding, attaining overall accuracies of 67.0% on CheXbench and 75.6% on Medical-CXR-VQA, and a RaTEScore of 0.569 on MIMIC-CXR for report generation. Code and data are available at this link.

News image captioning aims to produce journalistically informative descriptions by combining visual content with contextual cues from associated articles. Despite recent advances, existing methods struggle with three key challenges: (1) incomplete information coverage, (2) weak cross-modal alignment, and (3) suboptimal visual-entity grounding. To address these issues, we introduce MERGE, the first Multimodal Entity-aware Retrieval-augmented GEneration framework for news image captioning. MERGE constructs an entity-centric multimodal knowledge base (EMKB) that integrates textual, visual, and structured knowledge, enabling enriched background retrieval. It improves cross-modal alignment through a multistage hypothesis-caption strategy and enhances visual-entity matching via dynamic retrieval guided by image content. Extensive experiments on GoodNews and NYTimes800k show that MERGE significantly outperforms state-of-the-art baselines, with CIDEr gains of +6.84 and +1.16 in caption quality, and F1-score improvements of +4.14 and +2.64 in named entity recognition. Notably, MERGE also generalizes well to the unseen Visual News dataset, achieving +20.17 in CIDEr and +6.22 in F1-score, demonstrating strong robustness and domain adaptability.

Pretrained cross-modal models, such as the representative CLIP model, have recently led to a boom in the use of pretrained models for cross-modal zero-shot tasks due to their strong generalization abilities. However, we experimentally discovered that CLIP suffers from text-to-image retrieval hallucination, which adversely limits its capabilities under zero-shot learning. Specifically, in retrieval tasks, CLIP often assigns the highest score to an incorrect image, even when it correctly understands the image's semantic content in classification tasks. Accordingly, we propose the Balanced Score with Auxiliary Prompts (BSAP) method to address this problem. BSAP introduces auxiliary prompts that provide multiple reference outcomes for each image retrieval task. These outcomes, derived from the image and the target text, are normalized to compute a final similarity score, thereby reducing hallucinations. We further combine the original results with BSAP to generate a more robust hybrid outcome, termed BSAP-H. Extensive experiments on Referring Expression Comprehension (REC) and Referring Image Segmentation (RIS) tasks demonstrate that BSAP significantly improves the performance of CLIP and state-of-the-art vision-language models (VLMs). Code available at https://github.com/WangHanyao/BSAP.

This paper describes our method for Track 2 of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge on smoke-degraded images. In this task, smoke reduces image visibility and weakens the cross-view consistency required by scene optimization and rendering. We address this problem with a multi-stage pipeline consisting of image restoration, dehazing, MLLM-based enhancement, 3DGS-MCMC optimization, and averaging over repeated runs. The main purpose of the pipeline is to improve visibility before rendering while limiting scene-content changes across input views. Experimental results on the challenge benchmark show improved quantitative performance and better visual quality than the provided baselines. The code is available at https://github.com/plbbl/GenSmoke-GS. Our method achieved a ranking of 1 out of 14 participants in Track 2 of the NTIRE 3DRR Challenge, as reported on the official competition website: https://www.codabench.org/competitions/13993/#/results-tab.

Large Language Models (LLMs) encode vast world knowledge across multiple languages, yet their internal beliefs are often unevenly distributed across linguistic spaces. When external evidence contradicts these language-dependent memories, models encounter \emph{cross-lingual knowledge conflict}, a phenomenon largely unexplored beyond English-centric settings. We introduce \textbf{CLEAR}, a \textbf{C}ross-\textbf{L}ingual knowl\textbf{E}dge conflict ev\textbf{A}luation f\textbf{R}amework that systematically examines how multilingual LLMs reconcile conflicting internal beliefs and multilingual external evidence. CLEAR decomposes conflict resolution into four progressive scenarios, from multilingual parametric elicitation to competitive multi-source cross-lingual induction, and systematically evaluates model behavior across two complementary QA benchmarks with distinct task characteristics. We construct multilingual versions of ConflictQA and ConflictingQA covering 10 typologically diverse languages and evaluate six representative LLMs. Our experiments reveal a task-dependent decision dichotomy. In reasoning-intensive tasks, conflict resolution is dominated by language resource abundance, with high-resource languages exerting stronger persuasive power. In contrast, for entity-centric factual conflicts, linguistic affinity, not resource scale, becomes decisive, allowing low-resource but linguistically aligned languages to outperform distant high-resource ones.

Cross-modal retrieval facilitates more flexible information access and improves semantic understanding across different modalities. However, traditional cross-modal retrieval models rely on well-aligned datasets, which are often labor-intensive and costly to obtain. In real-world applications, data inevitably includes mismatched pairs, and these semantically inconsistent pairs can significantly degrade retrieval performance. Previous approaches have assumed ideal loss value distributions to optimize models for accurate semantic matching through soft-label estimation. However, the absence of hierarchical semantic correlation learning limits the effectiveness of these models in scenarios involving partial mismatches. To address these challenges, we propose Exploring Hierarchical Cross-Modal Correlation Consistency (EH3C) for cross-modal retrieval under partially mismatched conditions. Specifically, our approach first leverages neighborhood correlation distributions among samples to optimize cross-modal alignment, without assuming ideal distributions. This allows for the measurement of soft matching degrees between cross-modal data pairs and facilitates the effective learning of their positive correlations. Next, we enhance inter-class separability through intra-modal correlation learning by exploiting negative correlations between reliable negative sample pairs, thus enabling a more comprehensive exploration of cross-modal correlations. Finally, to assess the effectiveness and robustness of our approach, we conducted extensive experiments on three benchmark datasets. The results demonstrate that the proposed EH3C significantly improves cross-modal retrieval performance in scenarios involving partial mismatches.

Recent advancements in Large Language Models (LLMs)-based text embedding models primarily focus on data scaling or synthesis, yet limited exploration of training techniques and data quality, thereby constraining performance. In this work, we propose KaLM-Embedding-V2, a series of versatile and compact embedding models, systematically incentivizing advanced embedding capability in LLMs by superior training techniques and high-quality data. For model architecture, we implement the models on a 0.5B compact size with simple mean-pooling to produce fixed-length embeddings and remove the causal attention mask to enable fully bidirectional representation learning. For training techniques, we propose a progressive multi-stage training pipeline: pre-training on weakly supervised large-scale datasets, fine-tuning with supervised high-quality datasets, and contrastive distillation with fine-grained soft signals, integrated with focal-style reweighting and online hard-negative mixing to emphasize difficult samples and enrich hard negatives, respectively. For training data, we curate over 20 categories for pre-training and 100 categories for fine-tuning and contrastive distillation, to improve both performance and generalization, leveraging task-specific instructions, hard-negative mining, and example-based multi-class labeling to ensure high quality. Combining these techniques, our KaLM-Embedding-V2 series achieves state-of-the-art performance on the Massive Text Embedding Benchmark, outperforming models of comparable size and rivaling models 3-26x larger, setting a new standard for versatile and compact embedding models under 1B parameters. The code, data, and models will be publicly available to facilitate academic research.

Vision models for embodied intelligence require efficient 3D comprehension and interaction with objects within the scene. Existing 3D reconstruction models either overlook instance-level perception or rely on time-consuming offline reasoning, showing a less adaptability in real-time embodied scenario. In this paper, we present PromptDepth, the first promptable vision model that features both geometric 3D understanding and instance-level interaction especially designed for embodied intelligence. PromptDepth is a feed-forward network that quickly yields panoptic, instanced, or tracked depth map from two corresponding frames, enabling the real-time infer sequences from embodied agents. Specifically, following the minimal prediction problem, we design a promptable Dense Prediction Transformer, making it flexible to interact with unified dense prediction according to a specific prompt. Considering the substantial discrepancy between panoptic and instanced depth map, we further introduce a novel Instanced Label Distribution Smoothing (ILDS) loss, followed by Gram Anchoring, to mitigate the inherent conflict between dense and discrete representation. Trained on synthetic data only, our model achieves state-of-the-art results in both depth estimation and interactive segmentation on public benchmarks. Extensive experiments demonstrate superior visual efficiency in embodied tasks compared to current fundamental models. We believe that our efficient and flexible geometric 3D model offers a new foundation for vision tasks in embodied intelligence. The dataset and the code will be released.

Source-free domain adaptation (SFDA) is a practically significant and challenging problem. It aims to transfer the knowledge learned from a labeled source domain to an unlabeled target domain with a different data distribution while lacking access to the source domain. The target domain typically comprises two types of data: source-like and domain-specific data. However, identifying source-like data without access to the source domain remains a challenging task. Although some SFDA methods attempt to partition the target domain by using additional classifiers or relying on the prediction confidence of a single model, these approaches often suffer from suboptimal partitioning performance due to the potential biases of a single model towards the target domain. To address the above issue, we propose a novel dual-branch pseudo-label refinement method for SFDA. The proposed method employs a dual-branch framework to identify source-like data and divide the target domain data into two subsets. Following the divide-and-conquer principle, we design specific learning strategies tailored to each subset. We propose using the CLIPN model to identify erroneous samples in the source-like subset and correct them using the vision-language model (VLM). For the domain-specific subset, we introduce a novel prompt-learning technique based on the label space to optimize the label distribution. Both subsets utilize VLMs for pseudo-label refinement at varying degrees, enabling implicit alignment of label distributions. In experiments on three domain adaptation benchmark datasets, our method achieves the latest state-of-the-art results compared to existing approaches, with performance gains of +4.3% and +8.6% on the Office-Home and DomainNet datasets, respectively.

Depression ranks as one of the most prevalent psychological disorders and has received growing attention. Currently, there are two primary challenges in automatic video depression detection: the scarcity of labeled depressed data and effective feature representations characterizing human depression. To address these challenges, this work fully considers the correlation between human personality traits and depression, and proposes a Personality-guided Cross-domain Multitask Learning Network (PerDepNet) for automated video Depression detection. PerDepNet consists of a shared encoder for learning spatiotemporal joint representations, and a personality-guided decoder for cross-domain multitask learning tasks. The encoder contains a cross-scale shared feature extractor for capturing interactive spatial joint features at different scales, and a mamba-based temporal feature extractor for modeling long-term dynamics of depressed videos in personality and depression domains. The personality-guided decoder integrates the task of predicting personality traits into the depression detection task, enabling the personality traits to guide the estimate of depression levels in an adaptive weighted balance module. This is the first attempt to consider the role of personality traits in the depression detection task, thereby improving the performance of depression detection. Extensive experiments on three public depression video datasets demonstrate the superiority of the proposed method over state-of-the-art methods.

Multimodal Summarization (MMS) aims to generate concise textual summaries by understanding and integrating key information across multiple modalities such as videos, transcripts, and images. However, existing approaches still suffer from three main challenges: (1) reliance on domain-specific supervision, (2) implicit fusion with weak cross-modal grounding, and (3) flat temporal modeling without event transitions. To address these issues, we introduce CoE , a training-free MMS framework that performs structured reasoning through a Chain-of-Events guided by a Hierarchical Event Graph (HEG). The HEG encodes textual semantics into a structured prior that serves as a global scaffold for cross-modal reasoning.Guided by this hierarchy, CoE first performs cross-modal grounding to localize key visual cues, followed by event-evolution reasoning to capture temporal dependencies and causal transitions across the video.A lightweight style adaptation module further refines the generated summaries to match domain-specific linguistic conventions. Extensive experiments on eight diverse datasets demonstrate that CoE consistently outperforms state-of-the-art video CoT baselines, achieving average gains of +3.04 ROUGE , +9.51 CIDEr , and +1.88 BERTScore , highlighting its robustness, interpretability, and superior cross-domain generalization.

Aerial object-goal navigation (Aerial ObjectNav) requires an Unmanned Aerial Vehicle (UAV) to navigate to target objects in large-scale outdoor environments using only visual observations and high-level object descriptions, without detailed step-by-step instructions. Existing approaches rely on local observations or short-term history, lacking comprehensive scene understanding and efficient spatial exploration strategies, which constrains their navigation capability in complex aerial scenarios. To address these challenges, we propose OctMem-Agent, an octree memory-augmented framework for aerial object-goal navigation. Specifically, we introduce an Adaptive Octree Memory that incrementally aggregates RGB-D observations into a hierarchical 3D representation, capturing both explored regions and unexplored frontiers across large-scale aerial environments. We further propose a Instruction-Guided Memory Query module that extracts task-relevant scene and exploration tokens through instruction-modulated queries. By integrating these tokens with visual observations and language instructions, OctoMem-Agent achieves comprehensive scene understanding and effective spatial exploration for target localization. Extensive experiments on the Aerial ObjectNav benchmark UAV-ON demonstrate that our method achieves a significant 7.5\% improvement in success rate over existing methods, validating the effectiveness of our design.