Yuchen Luo

Email: 592mcavoy@sjtu.edu.cn
Github: @592McAvoy

Biography

I am currenlty a postgraduate student at Shanghai Jiao Tong University (SJTU). I have joined SJTU-ThinkLab and supervised by Prof. Yan. I am currently working on Deepfakes Detection and Face Image Generation.

Education

Shanghai Jiao Tong University (SJTU) (Sep.2016 – Jun.2020)
  • Bachelor in Software Engineering(SE)
Shanghai Jiao Tong University (SJTU) (Sep.2020 – Present)
  • Master in Computer Science(CS)

Publication

CVPR2021 Generalizing Face Forgery Detection with High-frequency Features [paper]
Yuchen Luo, Yong Zhang, Junchi Yan, Wei Liu
  • Leverage both the color textures and the high-frequency noises to propose a new face forgery detector with better generalization ability.

Internship

Google, Shanghai, China (Jul.2019 – Sep.2019)
Software Engineering Intern, Amarna Team, Media Data Solutions Group
  • Extract structured features from crawled URLs.
  • Embrace AI technique to design a lightweight classifier to detect ephemeral URLs
  • Integrate learned model into existing service to reduce resource cost and redundant query burden.
Tencent AI Lab, Shenzhen, China (Oct.2019 – Jun.2020)
Research Intern, Computer Vision Center
  • GAN based human face synthesis and face swaping.
  • Improve the generalization ability of face forgery detection methods.
Tencent Youtu Lab, Shanghai, China (Jul.2021 – Dec.2021)
Research Intern
  • Utilize the StyleGAN2 model for high fidelity disentangled face image generation.
  • Disentangle the face identity and attributes for identity swapping and identity anonymization .

Project

Image-Based 3D Textured body Reasoning [github]
Advisor: Cewu Lu, Machine Vision and Intelligence Group, SJTU
  • Generate 3D human body with full texture and median-level details based on single photo input.
  • Conduct model rendering of RGB/Depth/Normal images.
  • Apply mean-value coordinate transformation in naked model to recover geometry details such as clothed wrinkles and hairs.
Model Retrieval Based On VR Stick Figures [github]
  • Developed a model retrieval system based on stick figure input from VR environment.
  • Use Fourier descriptors for model features extraction and matching.
Detailed Video Avatar [github]
  • Optimize 3D model details according to paper "Detailed human avatars from monocular video".
  • Taking advantage of OpenPose facial keypoints detection.
  • Use visual hull as guidance for geometry deformation.
Digit Classification and Model Analysis [github]
  • Investigate different techniques to solve the digit classification task.
  • Implemented four different classification models, including Logistic Regression, Kernel-based Logistic Regression, Convolutional Neural Network, and Gaussian Mixture Models.
Deep Learning and Feature Visualization [github]
  • Utilize Variational Auto-Encoder (VAE) for face image generation.
  • Investigate the effect to the generation ability of VAE w.r.t the learning rate, the weight of KL Divergence loss, the normalization function, and the latent dimension.
Search Engine for Wikipedia [github]
  • Build a search engine for Wikipedia with existing dumps.
  • Extract millions of pages and terms from the dumped data and constructed indexes.
  • Implement basic search and tolerant search (including spelling correction and wild-card search).
  • Utilize concurrent programming and champion list to speed up the search procedure。

Acknowledgement

css files come from shumochu. Thank you~