arXiv subjects

Yiran Shen

Publications and source records attributed to Yiran Shen.

At least 19 recordsLinked to original sources

arXiv2026

How Much Does It Cost to Answer My Question? Benchmarking Cloud VLM-based VQA Systems

Vision-language models (VLMs) are becoming a practical backend for mobile visual question answering (VQA) systems, enabling smartphones and smart glasses to answer users' questions about the physical world. Since modern VLMs remain difficult to run on mobile and edge devices, VQA systems increasingly offload inference to cloud-based VLMs. This gives mobile devices access to stronger computation, but it also makes visual input preparation a key system variable: how the image is prepared before offloading affects not only answer quality but also payload size, token cost, and system latency. Proprietary APIs expose little control over model internals or serving behavior, leaving client-side preprocessing as the main practical optimization space for downstream developers. Many such techniques have been proposed for visual offloading, yet their cost-quality impact on commercial cloud VLMs has never been studied. To fill this gap, we present VQABench, the first systematic benchmark that treats client-side input preprocessing as a controlled variable for cloud-VLM-based VQA. We evaluate 12 preprocessing techniques across three VQA datasets and four commercial VLMs from three providers, totaling 95,168 API calls. Our results show that preprocessing is not universally beneficial: its effectiveness depends on the target model, API paradigm, provider token-accounting rule, and task formulation. A poorly selected preprocessing strategy can increase deployment cost or latency while degrading answer accuracy. Overall, our benchmark clarifies when preprocessing helps, when it fails, and why, providing insights to guide future research and real-world deployment of VQA systems.

cs.CV↗

arXiv2026

Diffusion-guided optimization for full waveform inversion

We present a diffusion-guided full waveform inversion (FWI) study in which pretrained diffusion generative models are used as learned regularizers inside a PDE-constrained seismic inversion loop. We compare three training-free guidance strategies: Manifold-Preserving Guided Diffusion (MPGD), SDEdit-based initialization, and Split Gibbs Diffusion Sampling (SGDS), which alternates between FWI likelihood updates and diffusion-prior denoising. The proposed workflow keeps wave-equation modeling in the inversion loop and uses a geological prior to stabilize model components that are weakly constrained by the seismic data. Controlled GeoFWI experiments, benchmark-scale Marmousi and Overthrust tests, a difficult Sigsbee2A salt test, and noise-degradation studies show that SGDS improves reconstruction quality relative to conventional L2 and total-variation regularized FWI in clean and moderately noisy synthetic settings. Overall, these experiments demonstrate that diffusion-guided optimization can serve as a practical learned regularization strategy for synthetic FWI benchmarks while preserving the wave-equation modeling loop.

physics.geo-ph↗

arXiv2026

DynFOA: Generating First-Order Ambisonics with Conditional Diffusion for Dynamic and Acoustically Complex 360-Degree Videos

Spatial audio is crucial for immersive 360-degree video experiences, yet most 360-degree videos lack it due to the difficulty of capturing spatial audio during recording. Automatically generating spatial audio such as first-order ambisonics (FOA) from video therefore remains an important but challenging problem. In complex scenes, sound perception depends not only on sound source locations but also on scene geometry, materials, and dynamic interactions with the environment. However, existing approaches only rely on visual cues and fail to model dynamic sources and acoustic effects such as occlusion, reflections, and reverberation. To address these challenges, we propose DynFOA, a generative framework that synthesizes FOA from 360-degree videos by integrating dynamic scene reconstruction with conditional diffusion modeling. DynFOA analyzes the input video to detect and localize dynamic sound sources, estimate depth and semantics, and reconstruct scene geometry and materials using 3D Gaussian Splatting (3DGS). The reconstructed scene representation provides physically grounded features that capture acoustic interactions between sources, environment, and listener viewpoint. Conditioned on these features, a diffusion model generates spatial audio consistent with the scene dynamics and acoustic context. We introduce M2G-360, a dataset of 600 real-world clips divided into MoveSources, Multi-Source, and Geometry subsets for evaluating robustness under diverse conditions. Experiments show that DynFOA consistently outperforms existing methods in spatial accuracy, acoustic fidelity, distribution matching, and perceived immersive experience.

cs.SD↗

arXiv2026

UniRED: Unified RGB-D Video Frame Interpolation with Event Guidance

High frame-rate RGB-D videos are crucial for a variety of downstream tasks, including motion analysis, dynamic scene understanding, and 3D reconstruction. However, due to hardware and sensing constraints, practical RGB-D cameras are typically limited to low frame rates, making it difficult to capture rapid scene dynamics. Existing video interpolation methods have achieved strong performance on RGB data, but they are not readily applicable to RGB-D scenarios, where they often yield blurry boundaries, visible artifacts, and degraded geometric consistency. Furthermore, motion estimation from only two boundary frames is inherently under-constrained in complex dynamic scenes. Event cameras, by contrast, provide asynchronous measurements with ultra-high temporal resolution, offering dense motion cues. In this paper, we propose a unified multimodal framework for RGB-D video interpolation that jointly exploits RGB appearance, depth geometry, and event-based temporal cues. Specifically, it first extracts and fuses RGB, depth and event cues, then estimates bidirectional flow with motion basis refinement for RGB and Z-axial refinement for depth, and finally synthesizes the target RGB-D frame via bidirectional warping and soft blending. In addition, we construct a new RGB-D-Event dataset to alleviate the scarcity of tri-modal training data. Extensive experiments on a public benchmark and the proposed dataset demonstrate that our method achieves superior photometric fidelity for RGB interpolation and stronger geometric accuracy for depth interpolation than existing approaches.

cs.CV↗

arXiv2026

Pocket-Dentist: On-Device Dental Image Understanding via Efficient Multimodal Large Language Models

Evaluations of dental vision-language models remain fragmented across datasets, task definitions and metrics, and often ignore their computational cost. This limits their widespread deployment for dental screening outside specialist centres, where timely inference, limited hardware, and local handling of patient images are vital for practical, privacy-preserving clinical prescreening. Here we present Pocket-Dentist, an efficiency-aware benchmark for dental multimodal question answering that brings together three datasets spanning approximately 1,159 patients from BRAR and MetaDent, five task types and seven metrics. Across 14 typical VLMs, our results reveal an interesting observation: compact VLMs, such as 2B-parameter models, become competitive with much larger VLMs on most metrics after lightweight adaptation while requiring substantially lower computational costs in dental image understanding. Deployed locally on an iPhone 17 Pro, our finetuned compact VLM Pocket-Dentist-2B processed each sample in 4.31 s, reducing latency by 4.9x and memory use by 2.3x compared with a 7B baseline. Our project page is available at https://2026-icml.github.io/pocket-dentist-icml.

cs.CV↗

arXiv2026

Simultaneous Multi-objective Alignment Across Verifiable and Non-verifiable Rewards

Aligning large language models to human preferences is inherently multidimensional, yet most pipelines collapse heterogeneous signals into a single objective. We seek to answer what it would take to simultaneously align a model across various domains spanning those with: verifiable rewards, non-verifiable subjective preferences, and complex interactive scenarios. Such multi-objective alignment setups are often plagued by individual objectives being at odds with each other, resulting in inefficient training and limited user control during inference. To address these issues, we propose $\textbf{M}$ulti-$\textbf{A}$ction-$\textbf{H}$ead $\textbf{AL}$ignment with PRM-guided Dec$\textbf{O}$ding ($\textbf{MAHALO}$), a unified framework that standardizes PRM training across verifiable and non-verifiable settings for step-level supervision, performs vectorized multi-objective alignment with Multi-Action-Head DPO, and enables controllable inference through objective-specific weighting and PRM-guided decoding. Experiments across math reasoning, human values alignment, and multi-turn tutoring show that MAHALO jointly improves multiple objectives simultaneously with limited interference, while remaining generalizable and adaptable across domains and offering flexible user control at inference time. Our code is available at: https://github.com/pearls-lab/multiobj-align.

cs.LG↗

arXiv2026

SIGMA-ASL: Sensor-Integrated Multimodal Dataset for Sign Language Recognition

Automatic sign language recognition (SLR) has become a key enabler of inclusive human-computer interaction, fostering seamless communication between deaf individuals and hearing communities. Despite significant advances in multimodal learning, existing SLR research remains dominated by vision-based datasets, which are limited by sensitivity to lighting and occlusion, privacy concerns, and a lack of cross-modal diversity. To address these challenges, we introduce SIGMA-ASL, a large-scale multimodal dataset for SLR. The dataset integrates an Azure Kinect RGB-D camera, a millimeter-wave (mmWave) radar, and two wrist-worn inertial measurement units (IMUs) to capture complementary visual, radio-reflection, and kinematic information. Collected in a controlled studio environment with 20 participants performing 160 common American sign language (ASL) signs, SIGMA-ASL provides 93,545 temporally synchronized word-level multimodal clips. A unified sensing framework achieves millisecond-level alignment across modalities, enabling reliable sensor fusion and cross-modal learning. We further design standardized preprocessing pipelines and benchmarking protocols under both user-dependent and user-independent settings, offering a comprehensive foundation for evaluating single and multimodal SLR. Extensive experiments validate the dataset's quality and demonstrate its potential as a valuable resource for developing robust, privacy-preserving, and ubiquitous sign language recognition systems.

cs.HC↗

arXiv2026

DynFOA: Generating First-Order Ambisonics with Conditional Diffusion for Dynamic and Acoustically Complex 360-Degree Videos

cs.SD↗

arXiv2026

Less is More: Skim Transformer for Light Field Image Super-resolution

A light field image captures scenes through its micro-lens array, providing a rich representation that encompasses spatial and angular information. While this richness comes at significant data redundancy, most existing methods tend to indiscriminately utilize all the information from sub-aperture images (SAIs) in an attempt to harness every visual cue regardless of their disparity significance. However, this paradigm inevitably leads to disparity entanglement, a fundamental cause of inefficiency in light field image processing. To address this limitation, we introduce the Skim Transformer, a novel architecture inspired by the "less is more" philosophy. It features a multi-branch structure where each branch is dedicated to a specific disparity range by constructing its attention score matrix over a skimmed subset of SAIs, rather than all of them. Building upon it, we present SkimLFSR, an efficient yet powerful network for light field image super-resolution. Requiring only 67% of the prior leading method's parameters}, SkimLFSR achieves state-of-the-art results surpassing the best existing method by 0.63 dB and 0.35 dB PSNR at the 2x and 4x tasks, respectively. Through in-depth analyses, we reveal that SkimLFSR, guided by the predefined skimmed SAI sets as prior knowledge, demonstrates distinct disparity-aware behaviors in attending to visual cues. Last but not least, we conduct an experiment to validate SkimLFSR's generalizability across different angular resolutions, where it achieves competitive performance on a larger angular resolution without any retraining or major network modifications. These findings highlight its effectiveness and adaptability as a promising paradigm for light field image processing.

eess.IV↗

arXiv2025

Beyond Subspace Isolation: Many-to-Many Transformer for Light Field Image Super-resolution

The effective extraction of spatial-angular features plays a crucial role in light field image super-resolution (LFSR) tasks, and the introduction of convolution and Transformers leads to significant improvement in this area. Nevertheless, due to the large 4D data volume of light field images, many existing methods opted to decompose the data into a number of lower-dimensional subspaces and perform Transformers in each sub-space individually. As a side effect, these methods inadvertently restrict the self-attention mechanisms to a One-to-One scheme accessing only a limited subset of LF data, explicitly preventing comprehensive optimization on all spatial and angular cues. In this paper, we identify this limitation as subspace isolation and introduce a novel Many-to-Many Transformer (M2MT) to address it. M2MT aggregates angular information in the spatial subspace before performing the self-attention mechanism. It enables complete access to all information across all sub-aperture images (SAIs) in a light field image. Consequently, M2MT is enabled to comprehensively capture long-range correlation dependencies. With M2MT as the foundational component, we develop a simple yet effective M2MT network for LFSR. Our experimental results demonstrate that M2MT achieves state-of-the-art performance across various public datasets, and it offers a favorable balance between model performance and efficiency, yielding higher-quality LFSR results with substantially lower demand for memory and computation. We further conduct in-depth analysis using local attribution maps (LAM) to obtain visual interpretability, and the results validate that M2MT is empowered with a truly non-local context in both spatial and angular subspaces to mitigate subspace isolation and acquire effective spatial-angular representation.

eess.IV↗

arXiv2025

CreBench: Human-Aligned Creativity Evaluation from Idea to Process to Product

Human-defined creativity is highly abstract, posing a challenge for multimodal large language models (MLLMs) to comprehend and assess creativity that aligns with human judgments. The absence of an existing benchmark further exacerbates this dilemma. To this end, we propose CreBench, which consists of two key components: 1) an evaluation benchmark covering the multiple dimensions from creative idea to process to products; 2) CreMIT (Creativity Multimodal Instruction Tuning dataset), a multimodal creativity evaluation dataset, consisting of 2.2K diverse-sourced multimodal data, 79.2K human feedbacks and 4.7M multi-typed instructions. Specifically, to ensure MLLMs can handle diverse creativity-related queries, we prompt GPT to refine these human feedbacks to activate stronger creativity assessment capabilities. CreBench serves as a foundation for building MLLMs that understand human-aligned creativity. Based on the CreBench, we fine-tune open-source general MLLMs, resulting in CreExpert, a multimodal creativity evaluation expert model. Extensive experiments demonstrate that the proposed CreExpert models achieve significantly better alignment with human creativity evaluation compared to state-of-the-art MLLMs, including the most advanced GPT-4V and Gemini-Pro-Vision.

cs.AI↗

arXiv2025

Preference-Based Learning in Audio Applications: A Systematic Analysis

Despite the parallel challenges that audio and text domains face in evaluating generative model outputs, preference learning remains remarkably underexplored in audio applications. Through a PRISMA-guided systematic review of approximately 500 papers, we find that only 30 (6%) apply preference learning to audio tasks. Our analysis reveals a field in transition: pre-2021 works focused on emotion recognition using traditional ranking methods (rankSVM), while post-2021 studies have pivoted toward generation tasks employing modern RLHF frameworks. We identify three critical patterns: (1) the emergence of multi-dimensional evaluation strategies combining synthetic, automated, and human preferences; (2) inconsistent alignment between traditional metrics (WER, PESQ) and human judgments across different contexts; and (3) convergence on multi-stage training pipelines that combine reward signals. Our findings suggest that while preference learning shows promise for audio, particularly in capturing subjective qualities like naturalness and musicality, the field requires standardized benchmarks, higher-quality datasets, and systematic investigation of how temporal factors unique to audio impact preference learning frameworks.

cs.SD↗

arXiv2025

A Multi-Store Privacy Measurement of Virtual Reality App Ecosystem

Virtual Reality (VR) has gained increasing traction among various domains in recent years, with major companies such as Meta, Pico, and Microsoft launching their application stores to support third-party developers in releasing their applications (or simply apps). These apps offer rich functionality but inherently collect privacy-sensitive data, such as user biometrics, behaviors, and the surrounding environment. Nevertheless, there is still a lack of domain-specific regulations to govern the data handling of VR apps, resulting in significant variations in their privacy practices among app stores. In this work, we present the first comprehensive multi-store study of privacy practices in the current VR app ecosystem, covering a large-scale dataset involving 6,565 apps collected from five major app stores. We assess both declarative and behavioral privacy practices of VR apps, using a multi-faceted approach based on natural language processing, reverse engineering, and static analysis. Our assessment reveals significant privacy compliance issues across all stores, underscoring the premature status of privacy protection in this rapidly growing ecosystem. For instance, one third of apps fail to declare their use of sensitive data, and 21.5\% of apps neglect to provide valid privacy policies. Our work sheds light on the status quo of privacy protection within the VR app ecosystem for the first time. Our findings should raise an alert to VR app developers and users, and encourage store operators to implement stringent regulations on privacy compliance among VR apps.

cs.CR↗

arXiv2025

NVS-SQA: Exploring Self-Supervised Quality Representation Learning for Neurally Synthesized Scenes without References

Neural View Synthesis (NVS), such as NeRF and 3D Gaussian Splatting, effectively creates photorealistic scenes from sparse viewpoints, typically evaluated by quality assessment methods like PSNR, SSIM, and LPIPS. However, these full-reference methods, which compare synthesized views to reference views, may not fully capture the perceptual quality of neurally synthesized scenes (NSS), particularly due to the limited availability of dense reference views. Furthermore, the challenges in acquiring human perceptual labels hinder the creation of extensive labeled datasets, risking model overfitting and reduced generalizability. To address these issues, we propose NVS-SQA, a NSS quality assessment method to learn no-reference quality representations through self-supervision without reliance on human labels. Traditional self-supervised learning predominantly relies on the "same instance, similar representation" assumption and extensive datasets. However, given that these conditions do not apply in NSS quality assessment, we employ heuristic cues and quality scores as learning objectives, along with a specialized contrastive pair preparation process to improve the effectiveness and efficiency of learning. The results show that NVS-SQA outperforms 17 no-reference methods by a large margin (i.e., on average 109.5% in SRCC, 98.6% in PLCC, and 91.5% in KRCC over the second best) and even exceeds 16 full-reference methods across all evaluation metrics (i.e., 22.9% in SRCC, 19.1% in PLCC, and 18.6% in KRCC over the second best).

cs.CV↗

arXiv2025

In-context Ranking Preference Optimization

Recent developments in Direct Preference Optimization (DPO) allow large language models (LLMs) to function as implicit ranking models by maximizing the margin between preferred and non-preferred responses. In practice, user feedback on such lists typically involves identifying a few relevant items in context rather than providing detailed pairwise comparisons for every possible item pair. Moreover, many complex information retrieval tasks, such as conversational agents and summarization systems, critically depend on ranking the highest-quality outputs at the top, emphasizing the need to support natural and flexible forms of user feedback. To address the challenge of limited and sparse pairwise feedback in the in-context setting, we propose an In-context Ranking Preference Optimization (IRPO) framework that directly optimizes LLMs based on ranking lists constructed during inference. To further capture flexible forms of feedback, IRPO extends the DPO objective by incorporating both the relevance of items and their positions in the list. Modeling these aspects jointly is non-trivial, as ranking metrics are inherently discrete and non-differentiable, making direct optimization difficult. To overcome this, IRPO introduces a differentiable objective based on positional aggregation of pairwise item preferences, enabling effective gradient-based optimization of discrete ranking metrics. We further provide theoretical insights showing that IRPO (i) automatically emphasizes items with greater disagreement between the model and the reference ranking, and (ii) links its gradient to an importance sampling estimator, yielding an unbiased estimator with reduced variance. Empirical results show IRPO outperforms standard DPO approaches in ranking performance, highlighting its effectiveness in aligning LLMs with direct in-context ranking preferences.

cs.LG↗

arXiv2025

SAND: Boosting LLM Agents with Self-Taught Action Deliberation

Large Language Model (LLM) agents are commonly tuned with supervised finetuning on ReAct-style expert trajectories or preference optimization over pairwise rollouts. Most of these methods focus on imitating specific expert behaviors or promoting chosen reasoning thoughts and actions over rejected ones. However, without reasoning and comparing over alternatives actions, LLM agents finetuned with these methods may over-commit towards seemingly plausible but suboptimal actions due to limited action space exploration. To address this, in this paper we propose Self-taught ActioN Deliberation (SAND) framework, enabling LLM agents to explicitly deliberate over candidate actions before committing to one. To tackle the challenges of when and what to deliberate given large action space and step-level action evaluation, we incorporate self-consistency action sampling and execution-guided action critique to help synthesize step-wise action deliberation thoughts using the base model of the LLM agent. In an iterative manner, the deliberation trajectories are then used to finetune the LLM agent itself. Evaluating on two representative interactive agent tasks, SAND achieves an average 20% improvement over initial supervised finetuning and also outperforms state-of-the-art agent tuning approaches.

cs.CL↗

arXiv2025

Self-supervised Learning of Event-guided Video Frame Interpolation for Rolling Shutter Frames

Most consumer cameras use rolling shutter (RS) exposure, which often leads to distortions such as skew and jelly effects. These videos are further limited by bandwidth and frame rate constraints. In this paper, we explore the potential of event cameras, which offer high temporal resolution. We propose a framework to recover global shutter (GS) high-frame-rate videos without RS distortion by combining an RS camera and an event camera. Due to the lack of real-world datasets, our framework adopts a self-supervised strategy based on a displacement field, a dense 3D spatiotemporal representation of pixel motion during exposure. This enables mutual reconstruction between RS and GS frames and facilitates slow-motion recovery. We combine RS frames with the displacement field to generate GS frames, and integrate inverse mapping and RS frame warping for self-supervision. Experiments on four datasets show that our method removes distortion, reduces bandwidth usage by 94 percent, and achieves 16 ms per frame at 32x interpolation.

cs.CV↗

arXiv2025

A Survey on Personalized and Pluralistic Preference Alignment in Large Language Models

Personalized preference alignment for large language models (LLMs), the process of tailoring LLMs to individual users' preferences, is an emerging research direction spanning the area of NLP and personalization. In this survey, we present an analysis of works on personalized alignment and modeling for LLMs. We introduce a taxonomy of preference alignment techniques, including training time, inference time, and additionally, user-modeling based methods. We provide analysis and discussion on the strengths and limitations of each group of techniques and then cover evaluation, benchmarks, as well as open problems in the field.

cs.CL↗