arXiv subjects

Pengcheng Wan

Publications and source records attributed to Pengcheng Wan.

4 recordsLinked to original sources

arXiv2026

MixDiffusion: Mixing Diffusion-based Uni-condition Text-to-Image Generation Models for Multi-condition Image Synthesis

Recent advances in text-to-image (T2I) generation have enabled controllable image synthesis by incorporating conditions beyond text. However, most existing diffusion-based methods are limited to a single type of control condition (e.g., bounding boxes or keypoints), which restricts their flexibility. To address this limitation, we propose MixDiffusion, a training-free diffusion framework for multi-condition T2I generation. MixDiffusion theoretically supports an arbitrary number of control conditions, including bounding boxes, keypoints, sketches, depth maps, reference images, and text, by collaboratively integrating multiple pre-trained uni-condition diffusion models. The key insight of the proposed approach is to derive the predicted noise distribution in each denoising step of the diffusion-based multi-condition image generation model from the predicted noise distributions of multiple diffusion-based uni-condition models with a derived integration formula, which is supported by rigorous theory proof. Owing to its training-free nature, MixDiffusion is easy to deploy and readily extensible to new control modalities.

cs.CV↗

arXiv2025

Rapid general Electromagnetic Analysis with computational conformal geometry via Conformal Energy Minimization

We recently found that the electromagnetic scattering problem can be very fast in an approach expressing the fields in terms of orthonormal basis functions. In this paper we apply computational conformal geometry with the conformal energy minimization (CEM) algorithm to make possible fast solution of finite-frequency electromagnetic problems involving arbitrarily shaped, simply-connected metallic surfaces. The CEM algorithm computes conformal maps with minimal angular distortion, enabling the transformation of arbitrary simply-connected surfaces into a disk, where orthogonal basis functions can be defined and electromagnetic analysis can be significantly simplified. We demonstrate the effectiveness and efficiency of our method by investigating the resonance characteristics of two metallic surfaces: a square plate and a four-petal plate. Compared to traditional finite element methods (e.g., COMSOL), our approach achieves a three-order-of-magnitude improvement in computational efficiency, requiring only seconds to extract resonant frequencies and fields. Moreover, it reveals low-energy, doubly degenerate resonance modes that are elusive to conventional methods. These findings not only provide a powerful tool for analyzing electromagnetic fields on complex geometries but also pave the way for the design of high-performance electromagnetic devices.

physics.optics↗

arXiv2024

HDN: Hybrid Deep-Learning and Non-Line-of-Sight Reconstruction Framework for Transcranial Photoacoustic Imaging of Human Brain

Photoacoustic imaging combines the high contrast of optical imaging with the deep penetration depth of ultrasonic imaging, showing great potential in cerebrovascular disease detection. However, the ultrasonic wave suffers strong attenuation and multi-scattering when it passes through the skull tissue, resulting in the distortion of the collected photoacoustic signal. In this paper, inspired by the principles of deep learning and non-line-of-sight imaging, we propose an image reconstruction framework named HDN (Hybrid Deep-learning and Non-line-of-sight), which consists of the signal extraction part and difference utilization part. The signal extraction part is used to correct the distorted signal and reconstruct an initial image. The difference utilization part is used to make further use of the signal difference between the distorted signal and corrected signal, reconstructing the residual image between the initial image and the target image. The test results on a photoacoustic digital brain simulation dataset show that compared with the traditional method (delay-and-sum) and deep-learning-based method (UNet), the HDN achieved superior performance in both signal correction and image reconstruction. Specifically for the structural similarity index, the HDN reached 0.661 in imaging results, compared to 0.157 for the delay-and-sum method and 0.305 for the deep-learning-based method.

physics.med-ph↗

arXiv2022

Metasurface electron optics in graphene

For electron optics in graphene, the propagation effect has so far been the only physical mechanism available. The resulting electron-optics-based components are large in size and operate at low temperatures to avoid violating the ballistic transport limits. In this paper, Dirac fermion metasurfaces, electronic counterparts of optical metasurfaces, are introduced for graphene electronics. By a metasurface, formally a linear array of gate-bias-controlled circular quantum dots, the wavefront of electron beams can be shaped within a one-quantum-dot-diameter distance, far below the ballistic limits at room temperature. This provides opportunities to create electron-optics-based devices that operate under ambient conditions. Moreover, unlike optical metasurfaces, Dirac fermion metasurfaces have near-perfect operating efficiencies and their high tunability allows for free and fast switching among functionalities. The concept of metasurface electron optics might open up a promising avenue for improving the performance of quantum devices in Dirac fermion materials.

cond-mat.mes-hall↗