arXiv subjects

Jianfei Zhao

Publications and source records attributed to Jianfei Zhao.

8 recordsLinked to original sources

arXiv2026

DV-SFT: Direct Vision Supervision for Fine-Grained Visual Understanding

Multimodal large language models are typically trained end-to-end to predict ground-truth answers, yet supervision signals are applied exclusively to text tokens. Visual tokens, the core carriers of visual information, are optimized only implicitly as part of the context, leading to coarse-grained visual understanding. Prior works attempt to supervise visual inputs but inevitably rely on auxiliary components such as additional decoders or forward passes, because visual tokens lack readily interpretable labels. This limits their practical applicability. In this work, we propose \textbf{D}irect \textbf{V}ision \textbf{S}upervised \textbf{F}ine-\textbf{T}uning (DV-SFT), which constructs explicit, token-level supervision for visual tokens and trains them through the same next-token prediction objective used for text. Specifically, we exploit the direct vision--text correspondence in OCR-related scenarios and automatically label each visual token with the word in its corresponding image patch. DV-SFT treats the MLLM as a black box, requiring no architectural modifications or additional forward passes. Extensive experiments demonstrate the superiority of direct vision supervision. DV-SFT consistently outperforms standard SFT across three in-domain and four out-of-domain benchmarks. Further analyses show that vision supervision effectively enhances fine-grained visual understanding and achieves higher multimodal alignment efficiency.

cs.CV↗

arXiv2026

Revisiting Reinforcement Learning with Verifiable Rewards from a Contrastive Perspective

Group Relative Policy Optimization (GRPO) is one of the most widely adopted RLVR algorithms for post-training large language models on reasoning tasks. We first show that GRPO admits an equivalent discriminative reformulation, in which policy optimization maximizes the expected score gap between verified positive and negative rollouts. This reformulation reveals two objective-level limitations: likelihood-misaligned surrogate scores, in which clipped ratio-based scores are optimized rather than the sequence likelihoods that govern generation, and score-insensitive credit assignment, in which rollout-level credit does not reflect the current score gaps between positive and negative rollouts. To address these limitations, we propose ConSPO, a Contrastive Sequence-level Policy Optimization method that uses length-normalized sequence log-probabilities as rollout scores and contrasts verified positive rollouts against negative distractors within the same group. ConSPO optimizes a group-wise InfoNCE-style objective to adaptively strengthen updates for poorly separated positives and high-scoring negatives, together with a curriculum-scheduled margin that preserves separation pressure as training progresses. Experiments across diverse settings show that ConSPO outperforms strong baselines on challenging reasoning benchmarks.

cs.LG↗

arXiv2026

Tell Model Where to Look: Mitigating Hallucinations in MLLMs by Vision-Guided Attention

Visual attention serves as the primary mechanism through which MLLMs interpret visual information; however, its limited localization capability often leads to hallucinations. We observe that although MLLMs can accurately extract visual semantics from visual tokens, they fail to fully leverage this advantage during subsequent inference. To address this limitation, we propose Vision-Guided Attention (VGA), a training-free method that first constructs precise visual grounding by exploiting the semantic content of visual tokens, and then uses this grounding to guide the model's focus toward relevant visual regions. In image captioning, VGA further refines this guidance dynamically during generation by suppressing regions that have already been described. In VGA, each token undergoes only a single forward pass, introducing a negligible latency overhead. In addition, VGA is fully compatible with efficient attention implementations such as FlashAttention. Extensive experiments across diverse MLLMs and multiple hallucination benchmarks demonstrate that VGA achieves state-of-the-art dehallucination performance. Further analysis confirms that explicit visual guidance plays a crucial role in enhancing the visual understanding capabilities of MLLMs.

cs.CV↗

arXiv2025

ADHint: Adaptive Hints with Difficulty Priors for Reinforcement Learning

To address the limited capability expansion and low sample efficiency of Reinforcement Learning (RL), recent methods have integrated ''hints'' into post-training, which are prefix segments of complete reasoning trajectories, aiming for powerful knowledge expansion and reasoning generalization. However, existing hint-based RL methods often neglect the role of difficulty in the hint-ratio schedule and relative-advantage estimation, resulting in unstable learning and excessive imitation of off-policy hints. To address this, we propose ADHint, which explicitly integrates difficulty into both processes to achieve a better trade-off between exploration and imitation. Specifically, we propose Adaptive Hint with Sample Difficulty Prior, which evaluates the difficulty of each sample under the current policy to schedule an appropriate hint ratio for rollout generation. Furthermore, we introduce Consistency-based Gradient Modulation alongside Selective Masking for Hint Preservation, which jointly modulate token-level gradients within hints to prevent biased and destructive updates. Additionally, we propose Advantage Estimation with Rollout Difficulty Posterior, which leverages the relative difficulty of rollouts with and without hints to compute their respective advantages, yielding more balanced updates. Extensive experiments across diverse modalities, scales, model families, and domains show that ADHint achieves superior reasoning capabilities and out-of-distribution generalization. Code will be released upon paper acceptance.

cs.CV↗

arXiv2025

Cross-Layer Vision Smoothing: Enhancing Visual Understanding via Sustained Focus on Key Objects in Large Vision-Language Models

Large Vision-Language Models (LVLMs) can accurately locate key objects in images, yet their attention to these objects tends to be very brief. Motivated by the hypothesis that sustained focus on key objects can improve LVLMs' visual capabilities, we propose Cross-Layer Vision Smoothing (CLVS). The core idea of CLVS is to incorporate a vision memory that smooths the attention distribution across layers. Specifically, we initialize this vision memory with position-unbiased visual attention in the first layer. In subsequent layers, the model's visual attention jointly considers the vision memory from previous layers, while the memory is updated iteratively, thereby maintaining smooth attention on key objects. Given that visual understanding primarily occurs in the early and middle layers of the model, we use uncertainty as an indicator of completed visual understanding and terminate the smoothing process accordingly. Experiments on four benchmarks across three LVLMs confirm the effectiveness and generalizability of our method. CLVS achieves state-of-the-art overall performance across a variety of visual understanding tasks and attains comparable results to the leading approaches on image captioning benchmarks.

cs.CV↗

arXiv2025

Mitigating Hallucination in Large Vision-Language Models through Aligning Attention Distribution to Information Flow

Due to the unidirectional masking mechanism, Decoder-Only models propagate information from left to right. LVLMs (Large Vision-Language Models) follow the same architecture, with visual information gradually integrated into semantic representations during forward propagation. Through systematic analysis, we observe that the majority of the visual information is absorbed into the semantic representations. However, the model's attention distribution does not exhibit sufficient emphasis on semantic representations. This misalignment between the attention distribution and the actual information flow undermines the model's visual understanding ability and contributes to hallucinations. To address this issue, we enhance the model's visual understanding by leveraging the core information embedded in semantic representations. Specifically, we identify attention heads that focus on core semantic representations based on their attention distributions. Then, through a two-stage optimization paradigm, we propagate the advantages of these attention heads across the entire model, aligning the attention distribution with the actual information flow. We evaluate our method on three image captioning benchmarks using five different LVLMs, demonstrating its effectiveness in significantly reducing hallucinations. Further experiments reveal a trade-off between reduced hallucinations and richer details. Notably, our method allows for manual adjustment of the model's conservativeness, enabling flexible control to meet diverse real-world requirements.

cs.CV↗

arXiv2025

Cross-Image Contrastive Decoding: Precise, Lossless Suppression of Language Priors in Large Vision-Language Models

Over-reliance on language priors is a major cause of hallucinations in Large Vision-Language Models (LVLMs), often leading to outputs that are linguistically plausible but visually inconsistent. Recent studies have explored contrastive decoding as a training-free solution. However, these methods typically construct contrastive visual inputs by perturbing the original image, resulting in distorted contrastive distributions, incomplete contrastive signals, and excessive suppression of language priors. Motivated by the observation that language priors tend to remain consistent across different images, we propose Cross-Image Contrastive Decoding (CICD), a simple yet effective training-free method that uses unrelated images as contrastive visual inputs. To address the issue of over-suppressing language priors, which can negatively affect the quality of generated responses, we further introduce a dynamic selection mechanism based on the cross-image differences in model behavior. By selectively suppressing language priors, our method reduces hallucinations without compromising the model's performance. Extensive experiments across multiple benchmarks and LVLMs confirm the effectiveness and generalizability of CICD, particularly in image captioning, where language priors are especially dominant.

cs.CV↗

arXiv2018

Opening the black box of deep learning

The great success of deep learning shows that its technology contains profound truth, and understanding its internal mechanism not only has important implications for the development of its technology and effective application in various fields, but also provides meaningful insights into the understanding of human brain mechanism. At present, most of the theoretical research on deep learning is based on mathematics. This dissertation proposes that the neural network of deep learning is a physical system, examines deep learning from three different perspectives: microscopic, macroscopic, and physical world views, answers multiple theoretical puzzles in deep learning by using physics principles. For example, from the perspective of quantum mechanics and statistical physics, this dissertation presents the calculation methods for convolution calculation, pooling, normalization, and Restricted Boltzmann Machine, as well as the selection of cost functions, explains why deep learning must be deep, what characteristics are learned in deep learning, why Convolutional Neural Networks do not have to be trained layer by layer, and the limitations of deep learning, etc., and proposes the theoretical direction and basis for the further development of deep learning now and in the future. The brilliance of physics flashes in deep learning, we try to establish the deep learning technology based on the scientific theory of physics.

cs.LG↗