Mengtian (Martin) Li

Perception at

Email: martinli [dot] work [at] gmail [dot] com

I was a Ph.D. student (2017-2022) at the Robotics Insitute of Carnegie Mellon University, where I was fortunate to work with Deva Ramanan. Previously, I was a master student at the same institute, advised by Daniel Huber. I received my B.S. from Kuang Yaming Honors School of Nanjing University.

My research interest is in computer vision and machine learning. In particular, I am interested in resource-constrained learning and inference.

News

May 2022 Migrated my personal website to github.io
Apr 2022 Defended my Ph.D. thesis! Thesis committee: Deva Ramanan, Martial Hebert, Mahadev (Satya) Satyanarayanan, Raquel Urtasun, and Ross Girshick.
Apr 2021 Invited talk at Georgia Tech RoboGrads Seminar
Mar 2021 Guest lecture at UIUC Advanced Computer Vision course
Feb 2021 Announcing the Streaming Perception Challenge (CVPR 2021)!
Oct 2020 Invited talk at Uber ATG
Aug 2020 Won the ECCV Best Paper Honorable Mention Award!
Apr 2020 Invited talk at Aurora

Papers

Neehar Peri, Mengtian Li, Benjamin Wilson, Yu-Xiong Wang, James Hays, Deva Ramanan. An Empirical Analysis of Range for 3D Object Detection. In ICCV Workshop, Oct 2023.

[Paper] [Bibtex]

Ziqi Pang, Deva Ramanan, Mengtian Li and Yu-Xiong Wang. Streaming Motion Forecasting for Autonomous Driving. In IROS, Oct 2023.

[Paper] [Code] [Bibtex]


Shengcao Cao, Mengtian Li, James Hays, Deva Ramanan, Yu-Xiong Wang and Liangyan Gui. Learning Lightweight Object Detectors via Progressive Knowledge Distillation. In ICML, Jul 2023.

[Paper] [Code] [Bibtex]


Chittesh Thavamani, Mengtian Li, Francesco Ferroni and Deva Ramanan. Learning to Zoom and Unzoom. In CVPR, Jun 2023.

[Project page] [Paper] [Talk] [Code] [Bibtex]

Shubham Gupta*, Jeet Kanjani*, Mengtian Li, Francesco Ferroni, James Hays, Deva Ramanan* and Shu Kong*. Far3Det: Towards Far-Field 3D Detection. In WACV, Jan 2023.

[Project page] [Paper] [Bibtex]

Neehar Peri, Jonathon Luiten, Mengtian Li, Aljosa Osep, Laura Leal-Taixé and Deva Ramanan. Forecasting from LiDAR via Future Object Detection. In CVPR, Jun 2022.

[Project page + code] [Paper] [Bibtex]

Xiaofang Wang, Shengcao Cao*, Mengtian Li*, Kris M. Kitani. Neighborhood-Aware Neural Architecture Search. In BMVC, 2021.

[Paper] [Bibtex]

Chittesh Thavamani*, Mengtian Li*, Nicolas Cebron, Deva Ramanan. FOVEA: Foveated Image Magnification for Autonomous Navigation. In ICCV, 2021.

[Project page] [Paper] [Poster] [Code] [Bibtex]

Mengtian Li, Yu-Xiong Wang and Deva Ramanan. Towards Streaming Perception. In ECCV, 2020.

Best Paper Honorable Mention

[Project page + talk + data] [Paper] [Code] [Bibtex]

Mengtian Li, Ersin Yumer and Deva Ramanan. Budgeted Training: Rethinking Deep Neural Network Training Under Resource Constraints. In ICLR, 2020.

[Project page] [Paper] [Talk] [Code] [Bibtex]

Mengtian Li, Zhe Lin, Radomír Měch, Ersin Yumer and Deva Ramanan. Photo-Sketching: Inferring Contour Drawings from Images. In WACV, 2019.

[Project page] [Paper] [Code] [Bibtex]

Mengtian Li, Laszlo Jeni, Deva Ramanan. Brute-Force Facial Landmark Analysis with a 140,000-Way Classifier. In AAAI, 2018.

[Paper] [Code] [Bibtex]

Mengtian Li, Daniel Huber. Guaranteed Parameter Estimation for Discrete Energy Minimization. In WACV, 2017.

[Paper] [Bibtex]

Mengtian Li, Alexander Shekhovtsov, Daniel Huber. Complexity of Discrete Energy Minimization Problems In ECCV, 2016.

Spotlight Presentation

[Paper] [Poster] [Talk] [Bibtex]

Misc

Confucius (孔子) once said: “if a craftsman wants to do good work, he must first sharpen his tools (工欲善其事，必先利其器).” I find that this concept also applies to research. Over the years, I have created various tools related to my research and I have some of them open sourced on Github:

  • HTML4Vision: a python-HTML-javascript tool for visualizing datasets, comparing algorithms and making figures
  • MTCMon: a web-based cluster resource monitor (widely adopted at CMU RI)
  • parscript: utilities for parallel or distributed execution of jobs
  • nntime: utilities for timing deep network modules
  • llcv: an extensible framework for low-latency computer vision research
  • DLGPUBench: a latency-focused GPU benchmark for deep learning