Resume Yuki Tsuji

Portfolio

Monocular 3D localization using 3D LiDAR Maps

Master thesis project: using ROS, PCL, OpenCV, Visual Odoemtry, g2o, OpenMP

・Matching visual odometry results and 3D LiDAR map

・Graph based optimiation will improve localization results

・Run about 60ms per frame

　Commint Soon ...

VoxelNet implementation from scratch

Voxelnet implementation in Chainer, ROS

Original paper is "VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection "

・Hash based efficient input creater is implemented

・Visualizer using ROS is implemented

・Multi-gpu training, lr scheduler are implemented

　[github] [paper]

Improved beyond pixel tracker in C++

Improved beyond pixel tracker in C++, OpenCV

Original paper is "Beyond Pixels: Leveraging Geometry and Shape Cues for Online Multi-Object Tracking"

・3D-3D/3D-2D IOU matching algorithm is implemented

・Run about 2ms per frame

・The code will be integrated into Autoware

　[movie] [paper]

Autoware -Opensource Committer-

Autoware is a open source software for autonomous driving systems

・developed 3D object tracking system using beyond pixel tracker

・developed Rosbag data extractor using ROS, OpenCV, PCL

・developing 3D object detection system using VoxelNet

　[movie] [code]

Yolov2 implementation from scratch

Yolov2 implementation in Chainer, ChainerCV, Python3

Original paper is "YOLO9000: Better, Faster, Stronger"

・Reproduce the results of the original paper

・Faster than original implementation in darknet

・Multi-gpu/scale training, lr scheduler are implemented

　[github] [paper]

SfMLearner implementation from scrach

SfMLearner implementation in Chainer, Python3

Original paper is "Unsupervised Learning of Depth and Ego-Motion from Video"

・Reproduce the results of the original paper

・Evaluate depth and odometry results by KiTTI Dataset

・Multi-gpu training, lr scheduler are implemented

　[github] [paper]

ENet implementation in Chainer

ENet implementation in Chainer, ChainerCV, Python3

Original paper is "ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation"

・Reproduce the results of the original paper

・Evaluate segmentation results by Cityscapes Dataset

・Class weighting and spatial dropout are implemented

　[github] [paper]

3D FCN for Vehicle Detection in TensorFlow

3D FCN for Vehicle Detection in Point Cloud by Tensorflow, ROS, Python

Original paper is "3D Fully Convolutional Network for Vehicle Detection in Point Cloud"

・LIDAR based 3D convolutional network

・Visualizer using ROS is implemented

・Dirty code...

　[github] [paper]

Experience

Preferred Networks, Inc.

Summer Internship and Part-Time Engineer

Developing deep-learning based algorithms for computer vision

Aug 2017 - Present　

Tier Ⅳ, Inc.

Part-Time Software Engineer

Developing multi-object tracking, SLAM and localization systems for autonomous driving systems

Apr 2017 - Present　

Grid, Inc.

Part-Time Engineer

Developed a deep learning framework 'ReNom', especially for the utilities of computer vision.

Jan 2017 - May 2017

Samurai Incubate, Inc.

Business Developer

Had an rich experience as a project manager of many hackathon events

Feb 2014 - May 2016

Portfolio

Monocular 3D localization using 3D LiDAR Maps

VoxelNet implementation from scratch

Improved beyond pixel tracker in C++

Autoware -Opensource Committer-

Yolov2 implementation from scratch

SfMLearner implementation from scrach

ENet implementation in Chainer

3D FCN for Vehicle Detection in TensorFlow

Experience

Preferred Networks, Inc.

Tier Ⅳ, Inc.

Grid, Inc.

Samurai Incubate, Inc.

Education

Tokyo University

Keio University

Publication

One paper accepted to IEEE ITSC2018