A5026686122- Face recognition and analysis
- Topological Materials and Phenomena
- Speech and Audio Processing
- Generative Adversarial Networks and Image Synthesis
- Big Data and Business Intelligence
- High-pressure geophysics and materials
- Advanced Condensed Matter Physics
- Medical Image Segmentation Techniques
- Cold Atom Physics and Bose-Einstein Condensates
- 2D Materials and Applications
- Advanced Manufacturing and Logistics Optimization
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Flexible and Reconfigurable Manufacturing Systems
University of Science and Technology of China
2025
The University of Texas at Dallas
2023
We report the design and observation of an ideal weak topological insulator in a quasi-one-dimensional bismuth halide, ${\mathrm{Bi}}_{4}{\mathrm{I}}_{1.2}{\mathrm{Br}}_{2.8}$. Via angle-resolved photoemission spectroscopy, we identify that ${\mathrm{Bi}}_{4}{\mathrm{I}}_{1.2}{\mathrm{Br}}_{2.8}$ hosts surface states on ${(100)}^{\ensuremath{'}}$ side form two anisotropic $\ensuremath{\pi}$-offset Dirac cones while topologically dark (001) top surface. The results fully determine unique...
Verifiable formal languages like Lean have profoundly impacted mathematical reasoning, particularly through the use of large language models (LLMs) for automated reasoning. A significant challenge in training LLMs these is lack parallel datasets that align natural with proofs. To address this challenge, paper introduces a novel framework translating Mathlib4 corpus (a unified library mathematics 4) into language. Building upon this, we employ dual augmentation strategy combines tactic-based...
Image Signal Processors (ISPs) convert raw sensor signals into digital images, which significantly influence the image quality and performance of downstream computer vision tasks. Designing ISP pipeline tuning parameters are two key steps for building an imaging system. To find optimal configurations, recent works use deep neural networks as a proxy to search or pipelines. However, these methods primarily designed maximize quality, sub-optimal in high-level tasks such detection, recognition,...
The paradigm of classifying three-dimensional (3D) topological insulators into strong and weak ones (STI WTI) opens the door for discovery various phases matter protected by different symmetries defined in dimensions. However, contrast to vast realization STIs, very few materials have been experimentally identified as being close WTI. Even amongst those identified, none exists with surface states (TSS) exposed a global bulk band gap that is stable at all temperatures. Here we report design...
Given one reference facial image and a piece of speech as input, talking head generation aims to synthesize realistic-looking video. However, generating lip-synchronized video with natural movements is challenging. The same clip can generate multiple possible lip movements, that is, there no one-to-one mapping relationship between them. To overcome this problem, we propose Speech Feature Extractor (SFE) based on memory-augmented self-supervised contrastive learning, which introduces the...
Given an audio clip and a reference face image, the goal of talking head generation is to generate high-fidelity video. Although some audio-driven methods generating videos have made achievements in past, most them only focused on lip synchronization lack ability reproduce facial expressions target person. To this end, we propose model consisting Memory-Sharing Emotion Feature extractor (MSEF) Attention-Augmented Translator based U-net (AATU). Firstly, MSEF can extract implicit emotional...