Zero-Shot Composed Image Retrieval (ZS-CIR) aims to retrieve images by combining a reference image with a textual description, without requiring supervision from target images. While existing self-supervised methods fuse visual and textual representations, they often struggle to capture fine-grained visual details and lack robust cross-modal interactions due to inadequate alignment supervision. To address this, we propose a novel framework, Dual-granularity Image-Text Alignment (DITA). DITA employs an Adaptive Token Pooling (ATP) module to compress visual tokens into an informative set, preserving rich image details for effective text querying while eliminating redundancy. It then introduces two core contrastive objectives: (i) an image-only prompt, formed by combining template text with the pooled tokens, aligns with the reference image for its cross-modal self-awareness, and (ii) a masked composite prompt, where a caption with masked keywords is appended to the image-only prompt, interacts with synthetic hard descriptions for image-text alignment and promotes their fine-grained fusion. To rigorously evaluate DITA, we conduct extensive experiments on FashionIQ, CIRR, and CIRCO benchmarks using diverse backbones (e.g., BLIP, BLIP-2, SigLIP). The experimental results demonstrate that DITA consistently outperforms competitive self-supervised baselines. Furthermore, DITA exhibits exceptional robustness against adversarial queries and varying text complexities. With both BLIP-based backbones, DITA processes each query in about 14 ms on a single RTX 4090 GPU. Compared with OSrCIR, a representative LLM-based training-free method relying on API calls, DITA achieves about 40x lower online latency while maintaining comparable retrieval performance. These results confirm that DITA provides a scalable, robust, and highly efficient solution for real-world ZS-CIR applications.
Projection stereolithography is a prominent 3D printing technology known for its simple design, high surface quality, and material compatibility. However, printing features near the nominal optical resolution remains challenging due to complex reasons, including Gaussian pixel radiance, nanoparticle scattering, and radical diffusion. Here, we developed a reaction‒diffusion–scattering model to predict and visualize the photopolymerization reaction with sub-pixel spatial resolution for photosensitive resin and resin–ceramic composites. Based on this model, we developed a segmentation exposure strategy combined with a grayscale mask compensation scheme, which enabled the fabrication of feature sizes below the pixel size of the mask. The model further enabled the continuous production of parts with a mere baseline top-down printing system via precise control of the photopolymerization process. Our study provides a promising avenue for the low-cost and high-speed fabrication of assembly-free multi-body components with tight tolerances and high resolution, including pillar bumps, gearsets, and microfluidic channels.
Introduction A substantial proportion of e-bike injuries and fatalities occur in motor-vehicle lanes, where riders are exposed to mixed traffic and higher crash risk, potentially offsetting the health benefits of e-bike use. Notably, many riders still enter motor-vehicle lanes even when dedicated bicycle lanes are available. Such lane intrusion increases exposure to hazardous interactions with vehicles and diminishes the safety benefits of cycling infrastructure. Therefore, identifying psychological factors associated with lane intrusion is important for informing injury-prevention interventions. Methodology An online survey of 729 e-bike riders (650 valid for analysis) assessed self-reported e-bike lane intrusion behaviour (e-bike LIB) using a newly developed Riding Lane Intrusion Behaviour Questionnaire, and assessed infrastructure perception, perceived safety, perceived risk, and Theory of Planned Behaviour constructs. We combined multi-group structural equation modelling with explainable machine learning to examine theory-informed associations among infrastructure perception, key psychological constructs, and e-bike LIB, and to explore potential non-linear patterns. Results Attitude, perceived behavioural control, and subjective norms were positively associated with self-reported lane intrusion, with attitude showing the strongest association. Infrastructure perception was negatively associated with self-reported lane intrusion and exhibited a possible non-linear pattern, with larger marginal reductions at higher levels of infrastructure perception. As infrastructure perception increased from low levels, perceived safety increased while perceived risk decreased, and these perceptions were further associated with e-bike LIB primarily indirectly through attitude and perceived behavioural control. Conclusions E-bike LIB should be understood as a preventable mixed-traffic exposure rather than a neutral lane-choice decision. The findings suggest that infrastructure provision should be complemented by behavioural measures targeting favourable attitudes, overconfidence, and permissive social norms. In poorer riding environments, stricter enforcement and improvements to the perceived usability and visibility of existing facilities may help discourage non-compliance with available bicycle lanes.
To tackle the difficulties arising from the depletion of traditional energy sources, this paper designs and optimizes a novel solar absorber for clean energy applications, aiming to achieve high absorption rate, ultra-broadband response, and polarization insensitivity. The proposed absorber is composed of a titanium (Ti) substrate, an aluminum oxide (Al2O3) dielectric layer, and four sets of nested L-shaped Ti resonant arms. FEM simulation results indicate that the absorber delivers superior performance in the 300 – 4000 nm range. Specifically, an average absorption rate of 98.91% is reached over a 3700 nm bandwidth, and the entire band exhibits a minimum absorption exceeding 90%. When exposed to AM1.5 standard illumination and high‑temperature conditions, the absorber shows excellent performance in both solar absorption and thermal radiation. Moreover, the absorber maintains stable high absorption performance over an incident angle range of 0° – 60° and a polarization angle range of 0° – 90°, showing good adaptability to operating conditions. These results indicate that the proposed design offers significant advantages in absorption performance and exhibits great potential for clean energy utilization.
We employed numerical simulations to investigate the influence of the arrangement of the erbium-doped fiber, output coupler, and saturable absorber on pulse evolution in the passively mode-locked erbium-doped fiber laser. The results show that altering the intracavity component order reconstructs the spatial build-up of peak power and the accumulation process of nonlinear phase shifts, resulting in distinct mode-locking build-up dynamics and output characteristics. Among the investigated configurations, structures in which the pulse is first shaped by the saturable absorber and subsequently amplified by the erbium-doped fiber improve gain extraction efficiency and maintain high peak power over longer fiber segments, thereby enhancing the effective accumulation of nonlinear phase shifts. Compared with the other structures, this design exhibits superior performance in terms of pulse compression, peak output power, and time-bandwidth product, while maintaining a stable advantage under different gain conditions.
