State-of-the-art text-to-image (T2I) diffusion models often struggle to generate rare compositions of concepts, e.g., objects with unusual attributes. In this paper, we show that the compositional generation power of diffusion models on such rare concepts can be significantly enhanced by the Large Language Model (LLM) guidance. We start with empirical and theoretical analysis, demonstrating that exposing frequent concepts relevant to the target rare concepts during the diffusion sampling process yields more accurate concept composition. Based on this, we propose a training-free approach, R2F, that plans and executes the overall rare-to-frequent concept guidance throughout the diffusion inference by leveraging the abundant semantic knowledge in LLMs. Our framework is flexible across any pre-trained diffusion models and LLMs, and can be seamlessly integrated with the region-guided diffusion approaches. Extensive experiments on three datasets, including our newly proposed benchmark, RareBench, containing various prompts with rare compositions of concepts, R2F significantly surpasses existing models including SD3.0 and FLUX by up to 28.1%p in T2I alignment. Code is available at https://github.com/krafton-ai/Rare-to-Frequent.

The Internet of Things (IoT) is a modern means of communication and its integration with biosensing significantly enhanced the monitoring and sensing of human vital signs, with blood pressure (BP) emerging as a promising application within the healthcare IoT domain. This study introduces a real-time, machine learning (ML)-based 3-D printed wearable device for BP and heart rate (HR) estimation using an advanced bi-optical pulse-wave-velocity acquisition system (BOPAS), which employs a dual photoplethysmography with an IoT-integrated wireless architecture for remote healthcare monitoring. Unlike conventional approaches, the proposed architecture eliminates dependency on high-quality signals and computationally intensive ML models by directly demonstrating strong pulse wave velocity (PWV) and BP correlation (${r} =0.96$for systolic, 0.89 for diastolic). Moreover, it overcomes the limitations like pulse ejection period and real-time filtering delays found in existing PWV methods making it simpler, more accurate, and practical for real-time wearables. After evaluating 14 different ML models, support vector regressor (SVR) yielded the highest testing R-square (R2) scores, surpassing 0.9272 for systolic BP (SBP) and 0.7896 for diastolic BP (DBP). A trained SVR model is deployed on a custom made smartphone application and tested in real time, yielding a mean absolute error and standard deviation of$2.68\pm 1.61$mmHg for SBP,$2.94\pm 1.84$mmHg for DBP, and$0.81\pm 0.55$beats per min for HR. Additionally, 90.1% of SBP samples and 85.0% of DBP samples exhibit absolute error less than 5 mmHg, demonstrating a promising results compared to the other approach.