arXiv subjects

Shuoshuo Zhang

Publications and source records attributed to Shuoshuo Zhang.

10 recordsLinked to original sources

arXiv2026

Skim and Skip: Hierarchical Adaptive Inference for Efficient Multimodal Retrieval

Universal multimodal retrieval (UMR) increasingly adopts multimodal large language models (MLLMs) as unified embedding backbones, but their strong retrieval performance comes at substantial inference cost. Existing methods typically rely on uniformly dense inference, where all input tokens are processed through the entire model and matched using the final-layer [EOS] representation. However, this paradigm overlooks two key forms of heterogeneity in multimodal retrieval: token contributions to the final retrieval embedding are highly uneven, and different queries require markedly different amounts of inference depth. To address this, we propose Skim and Skip (SAS), a hierarchical adaptive inference framework for efficient multimodal retrieval. SAS first performs token-level evidence selection to preserve only the input information most relevant to the final retrieval embedding, and then performs depth-adaptive inference to determine whether the current representation is already sufficient for reliable matching. Experiments on 12 MMEB retrieval tasks show that SAS retains about 99% of the dense baseline's average retrieval performance while achieving up to 1.64 times end-to-end speedup and up to 66.3% FLOPs reduction.

cs.IR↗

arXiv2026

Ultrafast wide-field 3D topography with extended depth of field

Ultrafast optical imaging has enabled direct observation of femtosecond-nanosecond dynamics, yet three-dimensional (3D) dynamic measurements at high numerical aperture (NA) remain hindered by the intrinsically shallow depth of field (DoF) of conventional microscopes. Here, we propose an ultrafast, wide-field pump-probe interferometric microscope on a telecentric platform that significantly extends the effective DoF to ~18 micrometer at a high NA of 0.9 while maintaining high spatial resolution (down to 235 nm) and temporal resolution (~170 fs). The system enables single-frame 3D topography reconstruction without axial scanning or multi-view acquisition. We demonstrate these capabilities by capturing axial material flow during laser-induced microsphere melting that remain unobservable with conventional narrow-DoF systems, and by tracking the azimuthal rotation of ablation lobes during axial propagation of temporal focused spatiotemporal optical vortex (TF-STOV) pulses, directly revealing the spatiotemporal evolution of STOV-matter interactions

physics.optics↗

arXiv2026

Temporal Focusing Enables Distortion-Resistant high-intensity Spatiotemporal Optical Vortices

Spatiotemporal optical vortices (STOVs) carry transverse orbital angular momentum and offer new degrees of freedom for light-matter interactions. Yet conventional focusing of STOVs introduces spatiotemporal astigmatism: the beam diffracts while the pulse duration stays constant, causing the vortex to deform away from focus. Here we overcome this limitation by introducing spectral phase modulation into a temporal focusing configuration, where angular dispersion forces the pulse to compress only at the geometric focus so that the spatial and temporal dimensions focus and defocus together. Our approach generates stable STOVs with self-similar, distortion-free evolution over an extended focal region. Besides, the orbital angular momentum vector can be continuously steered from purely longitudinal to strongly tilted orientations by adjusting the spatial dispersion, objective focal length, or input beam size. More importantly, our method offers full compatibility with high NA focusing geometry, allowing high-intensity and high-resolution applications. We validate these properties through femtosecond laser ablation under high-NA conditions and interferometric spatiotemporal field reconstruction under low-NA conditions.

physics.optics↗

arXiv2026

See Less, See Right: Bi-directional Perceptual Shaping For Multimodal Reasoning

Large vision-language models (VLMs) often benefit from intermediate visual cues, either injected via external tools or generated as latent visual tokens during reasoning, but these mechanisms still overlook fine-grained visual evidence (e.g., polylines in charts), generalize poorly across domains, and incur high inference-time cost. In this paper, we propose Bi-directional Perceptual Shaping (BiPS), which transforms question-conditioned masked views into bidirectional where-to-look signals that shape perception during training. BiPS first applies a KL-consistency constraint between the original image and an evidence-preserving view that keeps only question-relevant regions, encouraging coarse but complete coverage of supporting pixels. It then applies a KL-separation constraint between the original and an evidence-ablated view where critical pixels are masked so the image no longer supports the original answer, discouraging text-only shortcuts (i.e., answering from text alone) and enforcing fine-grained visual reliance. Across eight benchmarks, BiPS boosts Qwen2.5-VL-7B by 8.2% on average and shows strong out-of-domain generalization to unseen datasets and image types.

cs.CV↗

arXiv2025

Synthetic Spatiotemporal Plasmonic Vortices On Chip

Spatiotemporal vortices are polychromatic modes that intertwine orbital angular momentum (OAM) in space and time. Here we introduce a new class of such vortices, spatiotemporal plasmonic vortices (STPVs), carrying nontrivial topological spin textures. They are generated by chronotopic interference of temporally delayed plasmonic eigen-vortices, where a $π$-phase dislocation in the space-frequency domain maps into a 2$π$ spiraling phase in space-time, with the resulting focus-defocus dynamics emulate U(1) gauge transitions. Using interferometric time-resolved photoemission electron microscopy (ITR-PEEM), we directly image their nanometer-attosecond (nano-atto) evolution and control vortex number and position. Quantum-path analysis of coherent two-photon photoemission (2PP) processes reveals the nonlinear plasmonic polarization fields and angular-momentum conservation, establishing STPVs as a platform for probing spatiotemporally structured quantum matter.

cond-mat.mes-hall↗

arXiv2025

PixelCraft: A Multi-Agent System for High-Fidelity Visual Reasoning on Structured Images

Structured images (e.g., charts and geometric diagrams) remain challenging for multimodal large language models (MLLMs), as perceptual slips can cascade into erroneous conclusions. Intermediate visual cues can steer reasoning; however, existing cue-based methods are constrained with low-fidelity image processing and linear, rigid reasoning patterns, limiting their effectiveness on complex structured-image tasks. In this paper, we propose PixelCraft, a novel multi-agent system for high-fidelity image processing and flexible visual reasoning on structured images. The system comprises a dispatcher, a planner, a reasoner, critics, and a set of visual tool agents. To achieve high-fidelity processing, we construct a high-quality corpus and fine-tune an MLLM into a grounding model, whose pixel-level localizations are integrated with traditional computer vision (CV) algorithms in tool agents. Building on this foundation, PixelCraft facilitates flexible visual reasoning through a dynamic three-stage workflow of tool selection, agent discussion, and self-criticism. Moreover, unlike prior linear reasoning patterns that simply append historical images, PixelCraft maintains an image memory to allow the planner to adaptively revisit earlier visual steps, explore alternative reasoning branches, and dynamically adjust the reasoning trajectory during discussion. Extensive experiments on challenging chart and geometry benchmarks demonstrate that PixelCraft significantly improves visual reasoning performance for advanced MLLMs, setting a new standard for structured image reasoning. Our code will be available at https://github.com/microsoft/PixelCraft.

cs.CV↗

arXiv2025

PeRL: Permutation-Enhanced Reinforcement Learning for Interleaved Vision-Language Reasoning

Inspired by the impressive reasoning capabilities demonstrated by reinforcement learning approaches like DeepSeek-R1, recent emerging research has begun exploring the use of reinforcement learning (RL) to enhance vision-language models (VLMs) for multimodal reasoning tasks. However, most existing multimodal reinforcement learning approaches remain limited to spatial reasoning within single-image contexts, yet still struggle to generalize to more complex and real-world scenarios involving multi-image positional reasoning, where understanding the relationships across images is crucial. To address this challenge, we propose a general reinforcement learning approach PeRL tailored for interleaved multimodal tasks, and a multi-stage strategy designed to enhance the exploration-exploitation trade-off, thereby improving learning efficiency and task performance. Specifically, we introduce permutation of image sequences to simulate varied positional relationships to explore more spatial and positional diversity. Furthermore, we design a rollout filtering mechanism for resampling to focus on trajectories that contribute most to learning optimal behaviors to exploit learned policies effectively. We evaluate our model on 5 widely-used multi-image benchmarks and 3 single-image benchmarks. Our experiments confirm that PeRL trained model consistently surpasses R1-related and interleaved VLM baselines by a large margin, achieving state-of-the-art performance on multi-image benchmarks, while preserving comparable performance on single-image tasks.

cs.CV↗

arXiv2025

Controllable perfect spatiotemporal optical vortices

Spatiotemporal optical vortices (STOVs), as a kind of structured light pulses carrying transverse orbital angular momentum (OAM), have recently attracted significant research interest due to their unique photonic properties. However, general STOV pulses typically exhibit an annular intensity profile in the spatiotemporal plane, with a radius that scales with the topological charge, limiting their potential in many applications. Here, to address this limitation, we introduce the concept of perfect spatiotemporal optical vortices (PSTOVs). Unlike STOV pulses, the intensity distribution of PSTOV wavepackets is nearly independent of the topological charge. We show that such wavepackets can be generated by applying the spatiotemporal Fourier transform to a Bessel-Gaussian mode in the spatiotemporal frequency domain. More importantly, the mode distribution of PSTOV wavepackets can be freely controlled by introducing azimuthal-dependent phase modulation, enabling conversion from a standard annular profile to arbitrary polygonal shapes. Finally, experimental results confirm the successful generation of these wavepackets. Our findings will expand the study of STOV pulses and explore their potential applications in optical communications, information processing, topological photonics, and ultrafast control of light-matter interactions.

physics.optics↗

arXiv2020

Realization of vortex pair and its application in optical tweezers

As one fundamental property of light, the orbital angular momentum (OAM) of photon has elicited widespread interest. Here, we theoretically demonstrate that the OAM conversion of light without spin state can occur in homogeneous and isotropic medium when a line-variant locally linearly polarized (LVLLP) beam is strongly focused by a high numerical aperture (NA) objective lens. The high-NA objective lens here acts as a modulator that enables the spin-to-orbital OAM conversion of the two components of left and right circular vibrations of the input SDLLP beam spatially separated. Hence, partial conversion from linear state to conjugate OAM states takes place, resulting in helical phases with opposite directions for the longitudinal component of the two foci. Furthermore, such customized light field can be used to identify and separate chiral objects in optical tweezers.

physics.optics↗

arXiv2018

Orbital angular momentum conversion of optical field without spin state

As one fundamental property of light, the orbital angular momentum (OAM) of photon has elicited widespread interest. Here, we theoretically demonstrate that the OAM conversion of light without any spin state can occur in homogeneous and isotropic medium when a specially tailored locally linearly polarized (STLLP) beam is strongly focused by a high numerical aperture (NA) objective lens. Through a high NA objective lens, the STLLP beams can generate identical twin foci with tunable distance between them controlled by input state of polarization. Such process admits partial OAM conversion from linear state to conjugate OAM states, giving rise to helical phases with opposite directions for each focus of the longitudinal component in the focal field.

physics.optics↗