arXiv subjects

Jiayao Liu

Publications and source records attributed to Jiayao Liu.

8 recordsLinked to original sources

arXiv2026

PSMP-CLIP: Patch-Prompt SAM and Multi-Semantic Prompting for CLIP-Based Zero-Shot Anomaly Detection

Zero-shot anomaly detection aims to localize anomalies without target-domain samples. Existing CLIP-based methods suffer from coarse anomaly maps and limited semantic prompts. We propose PSMP-CLIP, integrating patch-prompt SAM2 segmentation (PPSS) and multi-semantic guided prompt regularization (MSGPR). PPSS samples prompts directly from intermediate patch features, avoiding threshold drift and guiding SAM2 to produce precise masks. MSGPR uses multiple learnable prompts constrained by semantic anchors to preserve generalization. Experiments on 14 datasets show highly competitive performance, achieving the best pixel-level AUROC on MVTec AD, BTAD, DTD-Synthetic, CVC-ClinicDB, TN3K, Endo, and Kvasir.

cs.CV↗

arXiv2026

tcnerv:dual-domain temporal context modeling for implicit neural video compression

Video compression aims to minimize reconstruction distor tion under a constrained bit rate. Existing video implicit neural representations (INRs) often decode frames independently, leaving intermediate features unconditioned on previous reconstructions and content embeddings without explicit temporal prediction. We propose TCNeRV, which exploits reconstructed context in both feature and embedding domains. Its multi-scale temporal-context fusion (MTCF) module injects gated historical features at multiple decoder scales, while temporal embedding-residual coding (TERC) predicts each content embedding and codes only its residual. With approximately 3M parameters, TCNeRV achieves an average PSNR of 36.08 dB on the UVG dataset, outperforming HNeRV-Boost by 2.20 dB. It reduces BD-rate by 22.06%, 66.73%, and 29.85% relative to HM, DCVC, and HiNeRV, respectively, demonstrating competitive rate-distortion performance with limited model capacity.

cs.CV↗

arXiv2026

JADE: Expert-Grounded Dynamic Evaluation for Open-Ended Professional Tasks

Evaluating agentic AI on open-ended professional tasks faces a fundamental dilemma between rigor and flexibility. Static rubrics provide rigorous, reproducible assessment but fail to accommodate diverse valid response strategies, while LLM-as-a-judge approaches adapt to individual responses yet suffer from instability and bias. Human experts address this dilemma by combining domain-grounded principles with dynamic, claim-level assessment. Inspired by this process, we propose JADE, a two-layer evaluation framework. Layer 1 encodes expert knowledge as a predefined set of evaluation skills, providing stable evaluation criteria. Layer 2 performs report-specific, claim-level evaluation to flexibly assess diverse reasoning strategies, with evidence-dependency gating to invalidate conclusions built on refuted claims. Experiments on BizBench show that JADE improves evaluation stability and reveals critical agent failure modes missed by holistic LLM-based evaluators. We further demonstrate strong alignment with expert-authored rubrics and effective transfer to HealthBench and DR.BENCH, covering medical and 10-domain professional evaluation settings. Code and data are available at https://github.com/smiling-world/JADE.

cs.AI↗

arXiv2026

GeoAgentBench: A Dynamic Execution Benchmark for Tool-Augmented Agents in Spatial Analysis

The integration of Large Language Models (LLMs) into Geographic Information Systems (GIS) marks a paradigm shift toward autonomous spatial analysis. However, evaluating these LLM-based agents remains challenging due to the complex, multi-step nature of geospatial workflows. Existing benchmarks primarily rely on static text or code matching, neglecting dynamic runtime feedback and the multimodal nature of spatial outputs. To address this gap, we introduce GeoAgentBench (GABench), a dynamic and interactive evaluation benchmark tailored for tool-augmented GIS agents. GABench provides a realistic execution sandbox integrating 117 atomic GIS tools, encompassing 53 typical spatial analysis tasks across 6 core GIS domains. Recognizing that precise parameter configuration is the primary determinant of execution success in dynamic GIS environments, we designed the Parameter Execution Accuracy (PEA) metric, which utilizes a "Last-Attempt Alignment" strategy to quantify the fidelity of implicit parameter inference. Complementing this, a Vision-Language Model (VLM) based verification is proposed to assess data-spatial accuracy and cartographic style adherence. Furthermore, to address the frequent task failures caused by parameter misalignments and runtime anomalies, we developed a novel agent architecture, Plan-and-React, that mimics expert cognitive workflows by decoupling global orchestration from step-wise reactive execution. Extensive experiments with seven representative LLMs demonstrate that the Plan-and-React paradigm significantly outperforms traditional frameworks, achieving the optimal balance between logical rigor and execution robustness, particularly in multi-step reasoning and error recovery. Our findings highlight current capability boundaries and establish a robust standard for assessing and advancing the next generation of autonomous GeoAI.

cs.AI↗

arXiv2026

Reconfigurable Momentum-space vectorial lasing enabled by Quasi-BIC

Bound states in the continuum (BICs) have enabled lasers with rich momentum-space textures. However, the output patterns of quasi-BIC lasers remain largely static and confined to a few geometries. Here, a reconfigurable momentum-space vectorial laser was proposed based on two-dimensional photonic crystal. By selectively exciting quasi-BIC modes, we identify the geometric asymmetry factors favoring single BIC, dual-BIC, and radiative mode with BIC operation. This approach yields vectorial lasing with characteristic patterns lasing in momentum space of bidirectional double lobes (BDL), radially polarized ring with BDL, azimuthally polarized ring with BDL, and linearly polarized spot with BDL. Importantly, reversible switching between a single donut and a donut with BDL was achieved in the same device by varying the pump energy density. Our work establishes a compact, versatile platform for reconfigurable vectorial lasers, with potential applications in tunable optical tweezers, super-resolution imaging, and on-chip optical interconnects.

physics.optics↗

arXiv2026

Reconfigurable and Recyclable Low-Threshold Quasi-BIC Lasers via a Tunable polymer Coating

Reconfigurable and sustainable microcavity lasers are highly desirable for next-generation integrated photonics. Here, we report a recyclable, low-threshold quasi-bound state in the continuum (q-BIC) laser fabricated via low-cost, high-throughput interference lithography. By introducing a polyvinyl alcohol (PVA) coating on a dye-doped photonic crystal, we suppress out-of-plane symmetry breaking, which reinforces optical confinement and reduces the lasing threshold. The q-BIC modes are further tuned through tailoring the refractive-index of the PVA layer by using Kramers-Kronig relation via Rhodamine 6G doping, demonstrating a wavelength shift of 7.14 nm and a sensitivity of 215 nm RIU as a sensing prob. More importantly, lasing modes are reversibly tuning via precisely controlling the coating thickness. Exploiting the dissolving and re-coating process, the laser is repeatedly reconfigured while maintaining performance. This work provides a sustainable and adaptive platform for sensing and reconfigurable photonic systems.

physics.optics↗

arXiv2026

Spatiotemporal Topological Phase Transition in non-Hermitian Photonic System

While energy band topology in spatial photonic crystals (PCs) and momentum-band topology in temporal crystals have each served as powerful probes of topological phases in their respective domains, their unification in a static platform remains unexplored. In this Letter, we bridge this gap by introducing a waveguide assisted non-Hermitian SSH model, in which controlled tuning of loss and coupling drives PT-symmetry breaking and enables a continuous transition between energy- and momentum-gap regimes. This allows us to construct a complete spatiotemporal topological phase diagram in a unified parameter space. By mapping this phase diagram onto a spatially graded PC, we experimentally observe multiple Bloch momentum-band gaps and a continuous spatiotemporal topological transition via translating across the static sample, enabling real-time control over the evolution pathway of the band topology. Our work creates a versatile, bias-free platform for exploring synthetic spacetime physics and opens new avenues for controlling light via non-Hermitian band engineering.

physics.optics↗

arXiv2026

Dual Flat-Bands of Bound State in the Continuum and Radiative Mode via TE-TM Coupling

A general symmetry-controlled mechanism is proposed for realizing dual flat-bands of bound state in the continuum (BIC) and its radiative counterpart in photonic crystal slabs. By breaking the vertical mirror symmetry of slab, inter-polarization coupling between TE-like and TM-like modes is activated, while intra-polarization coupling among modes within the same polarization class is simultaneously preserved. The cooperative action of these two coupling channels gives rise to the concurrent flattening of both the BIC-hosting band and the radiative band, resulting in a dual flat-band system with strongly contrasting quality (Q) factors. An effective two-step coupling model is constructed to capture the essential physics and show that the emergence of the flat bands is governed by geometric tuning rather than accidental degeneracies. The mechanism is shown to be generic with respect to polarization and material platform, enabling dual flat-band states in both low- and high-index systems, with substantially enhanced angular bandwidths in the latter. These finding establish a unified route for flat-band photonic engineering and provide a robust platform for angle-tolerant resonant photonic functionalities.

physics.optics↗