• Education
    • University of Maryland, College Park
      • Sep. 2015 – Dec.2018
      • Master in Electrical and Computer Engineering
    • Shanghai Jiao Tong Univeristy
      • Sep. 2011 – Jun. 2015
      • B.S. in Microelectronics
      • Top 1% B.S. Thesis Award
  • Publication
    • Kernel Foveated Rendering, Xiaoxu Meng, Ruofei Du, Matthias Zwicker, and Amitabh Varshney. Proc. ACM Comput. Graph. Interact. Tech., Vol. 1, No. 1, Article 5.
  • Research Areas & Skills
    • Languages: C++, Python, C#, Matlab, GLSL, HLSL
    • Research Topics: rendering acceleration, foveated rendering, deferred shading
    • Academic knowledge: computer graphics, machine learning, computer vision, random processes
  • Research Experience
    • Foveated Rendering
      • Designed a foveated rendering algorithm, which extends the classic log-polar transformation with kernel functions.
      • Implemented the kernel foveated rendering pipeline on the GPU, which has speedup of 4.8x for ray-marching
        scenes, and 2.7x for classic 3D meshes.
      • Conducted user study which explore the relationship between resolution and eccentricity, and redefine the visual
        acuity formula.
    • Foveated Path Tracing and Denoising
      • Designed a foveated path tracing and denoising system on FOVE HMD with eye tracking.
      • Implemented a 7-layer denoiser with edge-aware spatial wavelet filter and luminance variance filter, and generated same visual effect between 100 spp and 1spp rendering with speedup of 39x.
      • Constructed octree for mesh and used ordered traversal to facilitate ray-tracing for 5×10e4 x.
    • Denoising Videos with RNN Autoencoders
      • Implemented a convolutional autoencoder in Tensorflow with RNN for denoising rendered videos. The architecture has 8 convolutional encoding layers and 8 deconvolutional decoding layers. Each encoding block is a ”recurrent convolutional block” consisting of 3 convolutional layers.
      • Improved temporal stability of image sequences by spacial transformer.
    • Comprehensive Image Processing
      • Built edge detector with Leung-Malik, Schmid filter bank and improved precision of canny detector.
      • Implemented an end-to-end pipeline of panorama stitching of unordered images.
      • Created ”Play with Face” program for videos. The functions include face swap, average face, expression identifi-
        cation, and fancy filters.
      • Developed full pipeline of structure from motion (SfM) including two view reconstruction, triangulation, PnP,
        and bundle adjustment.
      • Constructed efficient and robust algorithms for 3D object reconstruction and segmentation.