Yuan Tong

About me

Education

Software Engineering (Bachelor) School of Computer Science, Wuhan University
2018.09 - 2022.06

Artificial Intelligence (Master) School of Computer Science, Wuhan University
2022.09 - 2024.06 (est.)

Experience

RD Internship at Video Infrastructure Department, ByteDance
05/2023 to now
  • Built several video enhancement and assessment model as production-ready module for the  BMF  framework used by the team, and participated in the framework development.
  • Developed CUDA Template library for pixel format conversion, supports conversion between various YUV and RGB formats with correct handling of colorspace. Achieves comparable or better performance compared with old implementation for single format.
  • Developed HEIF and AVIF encoding library using NVENC hardware acceleration, supporting 8-bit and 10-bit encoding and HDR metadata. Performed detail speed and quality benchmarking.
AVIF-Related Projects
12/2020 to now
  • Participated in the  libavif  project. Contributed features including progressive image encoding, changing encoding config, premultiply alpha support and higher precision colorspace conversion.
  • Participated in the  libyuv  project. Rewrote YUVToRGB implementation, developed specialized 2x bilinear scale up, extended HDR input support for various functions. Received Open Source Peer Bonus from Google.
  • Participated in the  libaom  project. Contributed  fixes  for a few bugs.
  • Contributed  to Firefox for support of full range videos and images using libyuv.
  • Developed  Chromatic  image server using Go and Angular. Supports automatic format conversion and delivery of JPEG, WEBP and AVIF formats and automatic compression quality decision. One instance has served over 100 million requests.
Applications of Neural Network
08/2022 to now
  • Developed  vs-NNVISR  VapourSynth plugin based on TensorRT, with an implementation of Deformable Convolution, supports applying multi-frame super-resolution and frame interpolation models.
  • Developed  TRT-NNScaler  tool based on TensorRT, supports applying enhancement and super-resolution models.
Backend RD Internship at Finance Department, ByteDance
06/2021 to 10/2021
  • Participated in architecture design of Business Rule Prediction System. Designed and implemented rule formalization, test-case generation, and test-case injection ability.
  • Designed static code analysis rules, and held introduction session to colleagues. Improved understanding of Golang language features and best practices for the whole team.
  • Designed and implemented timing trace and error aggregating utilities based on internal logging framework. Uncovered a few issues early and avoided potential incidents.
Network Tools
02/2021 to now
  • Developed remote network probing tool  Network Measure  using Golang. Implemented cross-platform ICMP packet process module, with Ping and MTR features based on it. Implemented DNS diagnose, TCP and TLS handshake test and speed test features. Supports remote call through HTTP and WebSocket.
  • Developed  smokeping-slave-go  compatible with the original version of SmokePing Master, while enabling collecting network metrics on more platforms.
Other Open Source Projects
  • Contributed x86 and ARM SIMD optimization code for  librcnb .
  • Maintainer of VapourSynth package on  conda-forge .
  • Developed  conda recipe  of TensorRT, to support install and dependency management through conda.