About me

Page
DSC03406_cropped_small

Yuanming Hu          胡渊鸣

I am a Ph.D. student at MIT CSAIL, advised by Frédo Durand and Bill Freeman.

I graduated with honor from Tsinghua University (Yao class) in July 2017. I worked on deep learning and computer vision, during my internship with Stephen Lin at Microsoft Research Asia. My undergraduate thesis is on automatic photo post-processing using reinforcement learning and adversarial learning (presented at SIGGRAPH 2018). I completed my master thesis (The ChainQueen Differentiable Physical Simulator) with Wojciech Matusik in November 2018. My work has been partly supported by an Edwin Webster fellowship and a Snap Research fellowship.

Contact:    yuanming _at_ mit.edu    Github

My current research interests include ...

  1. High-performance systems for computer graphics:
  2. Physical simulation for VFX, machine learning, and robotics:
    • ChainQueen: differentiable MLS-MPM simulator for soft robotics that enables skill learning in < 100 iterations gradient descent iterations;
    • MLS-MPM: efficient & simple MPM in 88 lines of code;
    • CPIC: rigid-deformable coupling and cutting for MPM;
    • AsyncMPM: MPM with adaptive regional time stepping.
  3. Computational photography and deep learning:
    • Exposure: learning infinite-resolution image processing with RL and GANs from unpaired images, based on a differentiable image editing model;
    • FC4: FCN-based color constancy with confidence-weighted pooling.

News

  • (07/2019) Welcome to our SIGGRAPH 2019 course On Hybrid Lagrangian-Eulerian Simulation Methods: Practical Notes and High-Performance Aspects (Thursday 3:45pm-5:15pm Room 408AB)! [Course Website]
  • (07/2019) We have one programming language paper conditionally accepted to SIGGRAPH Asia 2019.
  • (07/2019) MLS-MPM is featured on Two Minute Papers.
  • (06/2019) ChainQueen is on the MIT CSAIL news.
  • (05/2019) The code for our 1,040,875,347 voxel topology optimization is released.
  • (04/2019) We will present the course On Hybrid Lagrangian-Eulerian Simulation Methods: Practical Notes and High-Performance Aspects at SIGGRAPH 2019.
  • (03/2019) Our high-performance MLS-MPM with CPIC software (taichi_mpm) is now released under the commercially permissive MIT license (previously GPL).
  • (02/2019) I am selected as one of the 2019 Snap Research Fellows.
  • (01/2019) Our ChainQueen paper (differentiable MLS-MPM solver) is accepted by ICRA 2019  [Video].
  • (11/2018) My #include <taichi> talk introduces tricks I used to amalgamate Taichi into a single header.
  • (10/2018) Our 88-line MLS-MPM solver now supports all platforms with no external dependency. It has only 88 lines of essential code, real-time performance, and built-in UI for visualization. All these made it very suitable for learning and teaching MLS-MPM. [Visual result]
  • (09/2018) Our differentiable MLS-MPM solver for soft robotics is now on Arxiv  [Video].
  • (08/2018) Our high-performance MLS-MPM solver with CPIC is open source. MLS-MPM can be implemented within 88 lines of code.
  • (08/2018) We have two papers conditionally accepted to SIGGRAPH Asia 2018.
  • (07/2018) I will talk at GAMES: Graphics And Mixed Environment Seminar about our recent four projects (July 12 8:45pm: MLS-MPM-CPIC, AysncMPM [Slides]; July 19 8:45pm: Exposure, Taichi).
  • (06/2018) I will speak at the first event of Boston Computational Creativity group.
  • (06/2018) We have one simulation paper accepted to SCA 2018.
  • (05/2018) I will give the talk "From TensorFlow to Taichi: A GAN for Computational Photography, a Scheme for Physical Simulation, and a Library for Computer Graphics" at MIT CSAIL. Slides available upon request.
  • (04/2018) My article about building a reusable infrastructure for computer graphics is on arXiv.
  • (04/2018) We released our code for the CVPR 2017 paper "FC4 : Fully Convolutional Color Constancy with Confidence-weighted Pooling".
  • (03/2018) We have one simulation paper conditionally accepted to SIGGRAPH 2018.
  • (03/2018) The code for our Exposure project is now open-source.
  • (02/2018) Our paper Exposure: A White-Box Photo Post-Processing Framework is accepted by ACM Transactions on Graphics. We will present it at SIGGRAPH 2018.
  • ...

Research

2019

DiffTaichi: Differentiable Programming for Physical Simulation
arXiv 2019 [Code][Video]
Yuanming Hu
, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan
Ragan-Kelley, Fredo Durand

SIGGRAPH Asia 2019 [Code][Video]
Yuanming Hu, Tzu-Mao Li, Luke Anderson, Jonathan Ragan-Kelley, Fredo Durand

2018

ChainQueen: A Real-Time Differentiable Physical Simulator for Soft Robotics
ICRA 2019
Workshop on Modeling the Physical World: Perception, Learning, and Control (Oral), NIPS 2018
Yuanming Hu, Jiancheng Liu*, Andrew Spielberg* (joint 2nd authors), Joshua B. Tenenbaum, William T. Freeman, Jiajun Wu, Daniela RusWojciech Matusik
[Video] [Code]

Narrow-Band Topology Optimization on a Sparsely Populated Grid
ACM Transactions on Graphics, SIGGRAPH Asia 2018
Haixiang Liu*, Yuanming Hu* (joint first authors)Bo Zhu, Wojciech Matusik, Eftychios Sifakis
[Video] [Project Page & Code]

2017

☯An Asynchronous Material Point Method
SIGGRAPH 2017 Posters
Yuanming Hu*, Yu Fang* (joint first authors)
[Code]

Experience

Ph.D. Student@Massachusetts Institute of Technology 2017.9-
Right now I am a Ph.D. student at MIT CSAIL. I work on computational photography, physical simulation, and high-performance graphics systems.

mit

Visiting Student@The University of Pennsylvania 2017.7-2017.8
I visited Prof. Chenfanfu Jiang during summer 2017. We developed a new material point method that is faster and supports cutting and rigid-body coupling (SIGGRAPH 2018).

penn

Undergraduate Visiting Researcher@Stanford University 2016.6-2016.8
With Prof. Ron Fedkiw, I developed a novel line segment based reconstruction algorithm for fine tree details where previous methods do not perform well.

stanford

Internship Student@The University of Tokyo 2016.3-2016.6
Computer graphics research (physically based animation and rendering) with Prof. Seiichi Koshizuka and Prof. Toshiya Hachisuka. I implemented 10+ graphics papers during this internship and developed an initial version of Taichi, an open-source computer graphics library.

utokyo

Programming Contests

Asia-Pacific Informatics Olympiad 2012, China District

China National Olympiad in Informatics 2012

ACM-ICPC Asia Changsha Regional Contest 2013, with my team

ACM-ICPC Asia Shanghai Regional Contest 2014, with my team

China National Olympiad in Informatics 2014

Gold Medal, 1st / 450

Gold Medal, 13th / 350

Gold Medal, 3rd / 200

Gold Medal, 5th / 250

Problem Setter

Hobbies

Photography, graphic design, animated films, creating and destroying everything in my physically based simulators