I am a PhD candidate at Michigan State University in CSE department. I am working in the Computer Vision Lab advised by Dr. Xiaoming Liu. I earned my bachelors in CS at Kettering University in Flint, MI where I concentrated in both data security and computer graphics (e.g., lightsabers).

My research interests are broadly in deep learning and computer vision topics. Most of my focus tends to involve object recognition in 2D or 3D space usually with camera setups. I also am interested in depth estimation, scene forecasting, etc. I am passionate about weak and self-supervised machine learning.

My hobbies include comics, board games, music, skateboarding, dogs etc.


Education


Research


GrooMeD-NMS: Grouped Mathematically Differentiable NMS for Monocular 3D Object Detection
Abhinav Kumar, Garrick Brazil, Xiaoming Liu
Computer Vision and Pattern Recognition (CVPR 2021), Virtual, Jun. 2021
Bibtex |  arXiv |  PDF |  Supplemental |  Project Website |  Code

Kinematic 3D Object Detection in Monocular Video
Garrick Brazil, Gerard Pons-Moll, Xiaoming Liu, Bernt Schiele
European Conference on Computer Vision (ECCV 2020), Virtual, Aug. 2020
Bibtex |  arXiv |  PDF |  Supplemental |  Project Website |  Code

The Edge of Depth: Explicit Constraints between Segmentation and Depth
Shengjie Zhu, Garrick Brazil, Xiaoming Liu
Computer Vision and Pattern Recognition (CVPR 2020), Seattle, Washington, Jun. 2020
Bibtex |  arXiv |  PDF |  Supplemental |  Project Website |  Code

M3D-RPN: Monocular 3D Region Proposal Network for Object Detection
Garrick Brazil, Xiaoming Liu
International Conference on Computer Vision (ICCV 2019), Seoul, South Korea, Oct. 2019 (Oral 4.3%)
Bibtex |  arXiv |  PDF |  Project Website |  Code

Pedestrian Detection with Autoregressive Network Phases
Garrick Brazil, Xiaoming Liu
Computer Vision and Pattern Recognition (CVPR 2019), Long Beach, California, Jun. 2019
Bibtex |  arXiv |  PDF |  Project Website |  Code

Recurrent Flow-Guided Semantic Forecasting
Adam M. Terwilliger, Garrick Brazil, Xiaoming Liu
Winter Conference on Application of Computer Vision (WACV 2019), Waikoloa, Hawaii, Jan. 2019
Bibtex |  arXiv |  PDF |  Project Website |  Code

Illuminating Pedestrians via Simultaneous Detection & Segmentation
Garrick Brazil, Xi Yin, Xiaoming Liu
International Conference on Computer Vision (ICCV 2017), Venice, Italy, Oct. 2017
Bibtex |  arXiv |  PDF |  Project Website |  Code

Side Projects

These are a collection of projects which I conducted over the years, which don't have any ties to publication or actual research.
Fooling CNNs
Implemented a GAN loss to generate residual images to augment urban driving scenes with 2 primary goals: photo realism and maximum confusion/anarchy for the pedestrian detection CNN. In essence, this project aimed to fool and attack a state-of-the-art detector. The effects of using the synthetic data for alternate training were also investigated. We find that state-of-the-art systems are not naturally robust to such attacks in that the miss-rate error will quadruple (4x) unless the network is trained directly on the real and synthetic image data.

Semi-synthetic Data
Built a proof-of-concept project for generating synthetic pedestrians. The project uses the MakeHuman tool to generate synthetic 3D pedestrian models with highly variable pose, shape, race, hair, clothing, etc. In addition to synthetic, we further use a simple cut-and-paste method based on pixel-level segmentation masks to generate real people. Finally, we place synthetic or real pedestrians onto arbitrary background images. We learn depth cues to place pedestrians at proper scales on the sidewalk or road regions only.

Multipi
Multipi is a simple collision-based game built in the Unity engine. It was inspired by an imaginary operation in mathematics that I call multipiecation. The only objective of the game is to throw pies at other pies. When pies collide they multiply in size or quantity depending on the circumstance. A man is in game creeps around eating pies, eventually multiplying himself. Score is kept in terms of pi and pie. Sadly the game is no longer supported in the chrome web player, but if you want to try it feel free to email and I can send over an old executable.

Lightsabers
This project was inspired by my friends and my desire to be Jedis. The project simulates up to two 3D Lightsabers from the user's webcam feed in real-time. The tool has in-built calibration to work with generic props/objects. It is a simple idea based around 2D detection of spheres and RGB thresholding. It was written in OpenCV and uses OpenGL to draw 3D Lightsabers ontop of the users webcam. We implemented random flickering and motion blur to give a simple glowing lightsaber effect. Code is released here.

Optical illusions
I've been fascinated by illusions for a long time, or anything which can confuse our minds. Here are a few particularly fun optical illusions generated and rendered by a computer. All of the illusions are drawn by hand using primitive OpenGL methods in C++, and programmed many ages ago for a graphics course while getting my bachelor degree. Although I am much more interested in why such renderings confuse the human brain, the project is still fun to look at. The relatively simple source code is released, and still available here.