arXiv subjects

Zhaoyang Wei

Publications and source records attributed to Zhaoyang Wei.

12 recordsLinked to original sources

arXiv2026

CALR: Continuous Anchored Latent Reasoning via Render-of-Thought Compression

Visual latent reasoning compresses rendered derivations into compact intermediate states, reducing textual reasoning overhead. Existing approaches differ in how they represent these states: continuous methods avoid vocabulary constraints, whereas discrete methods improve accuracy through quantization into a finite codebook. Our analysis of representative continuous and discrete systems identifies two functional requirements: answers must rely on latent states, and those states must carry valid, problem-specific reasoning. Continuous latents influence answers despite collapsed reasoning content, whereas discrete latents retain recoverable intermediate reasoning that answer prediction largely bypasses. To address these challenges, we propose Continuous Anchored Latent Reasoning (CALR), which connects latent formation with answer use through functional anchoring. With reference latents from information-balanced compression, CALR couples latent-mediated answer supervision with derivation-level semantic anchoring: the former routes answer supervision through intermediate states, while the latter grounds their decoded content in problem-specific derivations. A parallel-to-autoregressive curriculum develops sequential reasoning by conditioning subsequent latent blocks on generated prefixes. Evaluations on five mathematical reasoning benchmarks across model families show substantial accuracy gains. Under matched budgets, CALR gains 26.0 percentage points over a comparable continuous latent reasoning method. Further analyses show that its latents support answer prediction and carry problem-specific intermediate reasoning.

cs.CV↗

arXiv2026

Video-HolmesV2: Can MLLMs Reason with Spatio-Temporal Audio-Visual Evidence in Long Videos?

Multimodal Large Language Models have demonstrated impressive video understanding, yet their ability to reason over long-form narratives is often masked by visual-centric evaluations and inefficient context processing. Existing benchmarks over-rely on visual heuristics while marginalizing auditory cues, effectively reducing models to "silent observers" that bypass genuine cross-modal reasoning. Moreover, standard dense sampling creates an evidence-context trade-off: increasing frames to capture evidence inevitably leads to attention distraction and token explosion. To bridge these gaps, we present Video-HolmesV2, a novel benchmark designed for Deep Audio-Visual Coupling. Unlike previous works, it enforces an Evidence-Based Evaluation, requiring models to justify answers with precise spatio-temporal audio-visual evidence, thereby reducing confounding effects of guessing and hallucinated evidence. To support this, we introduce: (1) a Multi-Model Cross-Verification pipeline to ensure task rigor; (2) a Spatio-temporal Evidence-Aware Metric for fine-grained calibration. Furthermore, we propose an Audio-Text Guided Token Compression framework. By fusing task intent with auditory anchors, our method distills high-value reasoning cues to mitigate long-context noise. In our evaluation, even strong proprietary models achieve below 60% accuracy, while our approach outperforms comparable open-source omni-models.

cs.CV↗

arXiv2026

Evidence-Grounded Trustworthy Multimodal Reasoning and Evaluation Benchmark in Complex Urban Scenes

While Multimodal Large Language Models (MLLMs) demonstrate impressive performance in benign scenarios, their cognitive reliability deteriorates significantly in complex scenes under adverse conditions. In these settings, models often rely on implicit inference without sufficient visual evidence, leading to a disconnect between perception and reasoning. Meanwhile, existing outcome-oriented benchmarks evaluate only final predictions and fail to diagnose failures in the underlying reasoning process. To address this gap, the authors propose AD2-Bench, which introduces a Hierarchical Visual Diagnosis framework that decomposes reasoning into a structured Chain of Evidence (CoE). This fine-grained diagnosis reveals that robust multimodal reasoning fundamentally depends on accurate evidence acquisition. Building on this perspective, the authors formulate reasoning from a probabilistic viewpoint and identify two primary causes of reasoning failure: Spatial Ambiguity, where models fail to distinguish target objects from background clutter, resulting in localization errors; and Semantic Uncertainty, where degraded visual features lead to incorrect semantic interpretation, resulting in understanding errors. To overcome these evidence deficiencies, they further propose Evidence-grounded Visual Reasoning (EGVOR), which replaces implicit reasoning with the explicit generation of Evidence Atoms - structured spatial-semantic triplets that enforce tight alignment between localization and semantic understanding. The model is trained through a hierarchical curriculum that progresses from reflective supervision construction to reinforcement learning, where reducing reasoning variance is explicitly rewarded. Extensive experiments demonstrate that EGVOR substantially improves reasoning stability under adverse conditions, providing a more robust framework for trustworthy multimodal cognition.

cs.CV↗

arXiv2026

Evidence-Grounded Trustworthy Multimodal Reasoning and Evaluation Benchmark in Complex Urban Scenes

cs.CV↗

arXiv2026

Video Evidence to Reasoning Efficient Video Understanding via Explicit Evidence Grounding

Large Vision-Language Models (LVLMs) face a fundamental dilemma in video reasoning: they are caught between the prohibitive computational costs of verbose reasoning and the hallucination risks of efficient, ungrounded approaches. To resolve this, we introduce the Chain of Evidence (CoE), a novel framework that architecturally decouples and co-optimizes perceptual grounding and reasoning efficiency. CoE incorporates two core innovations: (1) A lightweight Evidence Grounding Module (EGM) that acts as a query-guided filter, dynamically identifying and extracting a compact set of high-fidelity visual evidence; and (2) An Evidence-Anchoring Protocol optimized via Reinforcement Learning. Crucially, we design a composite reward mechanism that enforces process alignment, compelling the model to strictly reference identified temporal anchors during deduction, thereby mitigating hallucinations. To enable this, we construct CoE-Instruct, a large-scale dataset (164k samples) featuring a novel dual-annotation schema for separate perception and reasoning supervision. Extensive experiments on five benchmarks, including Video-MME, MVBench, and VSI-Bench, demonstrate that CoE-enhanced models establish a new state-of-the-art. They significantly outperform existing methods in accuracy, proving CoE to be a powerful and practical paradigm for reliable video understanding.

cs.CV↗

arXiv2026

LUT: Latent Utility Training for Visual Reasoning

Multimodal large language models have advanced visual understanding, yet perception-intensive reasoning remains challenging. Recent latent visual reasoning methods introduce hidden-space computation before answering, but they often rely on costly intermediate supervision, such as bounding boxes, sketches, or interleaved rationales. These strategies focus on how latent states should be shaped, but do not explicitly assess whether the latent is useful for the final answer. We propose LUT, a latent reasoning framework trained with only standard VQA pairs. LUT centers training on Latent Utility at two levels. At the trajectory level, we propose Utility-Aware Latent Distillation SFT, which explores answer-relevant latent trajectories, selects qualified trajectories by their information gain, and distills more reliable and learnable supervision through curriculum learning. At the step level, we propose Latent Attribution Policy Optimization, which uses answer-to-latent attribution to differentially optimize latent steps during reinforcement learning. Experiments on perception-intensive visual reasoning benchmarks show that LUT outperforms previous latent reasoning methods and remains competitive with latent-text interleaved methods with lower annotation cost.

cs.CV↗

arXiv2026

What Makes VLMs Robust? Towards Reconciling Robustness and Accuracy in Vision-Language Models

Achieving adversarial robustness in Vision-Language Models (VLMs) inevitably compromises accuracy on clean data, presenting a long-standing and challenging trade-off. In this work, we revisit this trade-off by investigating a fundamental question: What makes VLMs robust? Through a detailed analysis of adversarially fine-tuned models, we examine how robustness mechanisms function internally and how they interact with clean accuracy. Our analysis reveals that adversarial robustness is not uniformly distributed across network depth. Instead, unexpectedly, it is primarily localized within the shallow layers, driven by a low-frequency spectral bias and input-insensitive attention patterns. Meanwhile, updates to the deep layers tend to undermine both clean accuracy and robust generalization. Motivated by these insights, we propose Adversarial Robustness Adaptation (R-Adapt), a simple yet effective framework that freezes all pre-trained weights and introduces minimal, insight-driven adaptations only in the initial layers. This design achieves an exceptional balance between adversarial robustness and clean accuracy. R-Adapt further supports training-free, model-guided, and data-driven paradigms, offering flexible pathways to seamlessly equip standard models with robustness. Extensive evaluations on 18 datasets and diverse tasks demonstrate our state-of-the-art performance under various attacks. Notably, R-Adapt generalizes efficiently to large vision-language models (e.g., LLaVA and Qwen-VL) to enhance their robustness. Our project page is available at https://summu77.github.io/R-Adapt.

cs.CV↗

arXiv2026

HeroGS: Hierarchical Guidance for Robust 3D Gaussian Splatting under Sparse Views

3D Gaussian Splatting (3DGS) has recently emerged as a promising approach in novel view synthesis, combining photorealistic rendering with real-time efficiency. However, its success heavily relies on dense camera coverage; under sparse-view conditions, insufficient supervision leads to irregular Gaussian distributions, characterized by globally sparse coverage, blurred background, and distorted high-frequency areas. To address this, we propose HeroGS, Hierarchical Guidance for Robust 3D Gaussian Splatting, a unified framework that establishes hierarchical guidance across the image, feature, and parameter levels. At the image level, sparse supervision is converted into pseudo-dense guidance, globally regularizing the Gaussian distributions and forming a consistent foundation for subsequent optimization. Building upon this, Feature-Adaptive Densification and Pruning (FADP) at the feature level leverages low-level features to refine high-frequency details and adaptively densifies Gaussians in background regions. The optimized distributions then support Co-Pruned Geometry Consistency (CPG) at parameter level, which guides geometric consistency through parameter freezing and co-pruning, effectively removing inconsistent splats. The hierarchical guidance strategy effectively constrains and optimizes the overall Gaussian distributions, thereby enhancing both structural fidelity and rendering quality. Extensive experiments demonstrate that HeroGS achieves high-fidelity reconstructions and consistently surpasses state-of-the-art baselines under sparse-view conditions.

cs.CV↗

arXiv2026

SAPNet++: Evolving Point-Prompted Instance Segmentation with Semantic and Spatial Awareness

Single-point annotation is increasingly prominent in visual tasks for labeling cost reduction. However, it challenges tasks requiring high precision, such as the point-prompted instance segmentation (PPIS) task, which aims to estimate precise masks using single-point prompts to train a segmentation network. Due to the constraints of point annotations, granularity ambiguity and boundary uncertainty arise the difficulty distinguishing between different levels of detail (eg. whole object vs. parts) and the challenge of precisely delineating object boundaries. Previous works have usually inherited the paradigm of mask generation along with proposal selection to achieve PPIS. However, proposal selection relies solely on category information, failing to resolve the ambiguity of different granularity. Furthermore, mask generators offer only finite discrete solutions that often deviate from actual masks, particularly at boundaries. To address these issues, we propose the Semantic-Aware Point-Prompted Instance Segmentation Network (SAPNet). It integrates Point Distance Guidance and Box Mining Strategy to tackle group and local issues caused by the point's granularity ambiguity. Additionally, we incorporate completeness scores within proposals to add spatial granularity awareness, enhancing multiple instance learning (MIL) in proposal selection termed S-MIL. The Multi-level Affinity Refinement conveys pixel and semantic clues, narrowing boundary uncertainty during mask refinement. These modules culminate in SAPNet++, mitigating point prompt's granularity ambiguity and boundary uncertainty and significantly improving segmentation performance. Extensive experiments on four challenging datasets validate the effectiveness of our methods, highlighting the potential to advance PPIS.

cs.CV↗

arXiv2025

Guiding the Inner Eye: A Framework for Hierarchical and Flexible Visual Grounded Reasoning

Models capable of "thinking with images" by dynamically grounding their reasoning in visual evidence represent a major leap in multimodal AI. However, replicating and advancing this ability is non-trivial, with current methods often trapped between the instability of end-to-end reinforcement learning (RL) and the rigidity of supervised fine-tuning (SFT). This leads to models that either struggle to learn or lack the cognitive flexibility required for complex, real-world scenes. To navigate this dilemma, we introduce GRiP (Guided Reasoning and Perception), a novel two-stage training framework that cultivates robust and flexible visual grounded reasoning by explicitly guiding the model's perceptual focus and logical pathways. GRiP's core lies in its cognitive-enhanced RL stage, which features two key innovations: (1) a Salience-Weighted IoU Reward that incentivizes the model to prioritize the localization of mission-critical objects over trivial distractors, and (2) a Multi-Heuristic Reward that encourages cognitive flexibility by rewarding diverse yet logically valid reasoning pathways. Initialized from the Qwen2.5-VL-7B model, GRiP demonstrates significant performance gains across multiple challenging benchmarks. It achieves state-of-the-art results among open-source models on the highly challenging TreeBench and V* Bench, proving its effectiveness in complex visual reasoning. Our work demonstrates that moving beyond simplistic rewards and instead guiding models with cognitively-inspired signals for what to see and how to think is crucial for unlocking the next level of multimodal intelligence. The code will be made publicly available.

cs.CV↗

arXiv2025

VER-Bench: Evaluating MLLMs on Reasoning with Fine-Grained Visual Evidence

With the rapid development of MLLMs, evaluating their visual capabilities has become increasingly crucial. Current benchmarks primarily fall into two main types: basic perception benchmarks, which focus on local details but lack deep reasoning (e.g., "what is in the image?"), and mainstream reasoning benchmarks, which concentrate on prominent image elements but may fail to assess subtle clues requiring intricate analysis. However, profound visual understanding and complex reasoning depend more on interpreting subtle, inconspicuous local details than on perceiving salient, macro-level objects. These details, though occupying minimal image area, often contain richer, more critical information for robust analysis. To bridge this gap, we introduce the VER-Bench, a novel framework to evaluate MLLMs' ability to: 1) identify fine-grained visual clues, often occupying on average just 0.25% of the image area; 2) integrate these clues with world knowledge for complex reasoning. Comprising 374 carefully designed questions across Geospatial, Temporal, Situational, Intent, System State, and Symbolic reasoning, each question in VER-Bench is accompanied by structured evidence: visual clues and question-related reasoning derived from them. VER-Bench reveals current models' limitations in extracting subtle visual evidence and constructing evidence-based arguments, highlighting the need to enhance models's capabilities in fine-grained visual evidence extraction, integration, and reasoning for genuine visual understanding and human-like analysis. Dataset and additional materials are available https://github.com/verbta/ACMMM-25-Materials.

cs.CV↗

arXiv2024

Semantic-aware SAM for Point-Prompted Instance Segmentation

Single-point annotation in visual tasks, with the goal of minimizing labelling costs, is becoming increasingly prominent in research. Recently, visual foundation models, such as Segment Anything (SAM), have gained widespread usage due to their robust zero-shot capabilities and exceptional annotation performance. However, SAM's class-agnostic output and high confidence in local segmentation introduce 'semantic ambiguity', posing a challenge for precise category-specific segmentation. In this paper, we introduce a cost-effective category-specific segmenter using SAM. To tackle this challenge, we have devised a Semantic-Aware Instance Segmentation Network (SAPNet) that integrates Multiple Instance Learning (MIL) with matching capability and SAM with point prompts. SAPNet strategically selects the most representative mask proposals generated by SAM to supervise segmentation, with a specific focus on object category information. Moreover, we introduce the Point Distance Guidance and Box Mining Strategy to mitigate inherent challenges: 'group' and 'local' issues in weakly supervised segmentation. These strategies serve to further enhance the overall segmentation performance. The experimental results on Pascal VOC and COCO demonstrate the promising performance of our proposed SAPNet, emphasizing its semantic matching capabilities and its potential to advance point-prompted instance segmentation. The code will be made publicly available.

cs.CV↗