Chen Feng in Chinese: 冯晨
ChenFeng's picture

Assistant Professor, Ph.D.

Department of Civil and Urban Engineering
Department of Mechanical and Aerospace Engineering
New York University Tandon School of Engineering
15 MetroTech Center, Brooklyn, NY

E-mail: cfeng[at]nyu[dot]edu
Work: 646-997-3445

Google Scholar LinkedIn GitHub

FoldingNet overview

FoldingNet: Point Cloud Auto-encoder via Deep Grid Deformation

The state-of-the-art unsupervised deep auto-encoder of point clouds which reconstruct order point clouds from unordered input, useful for autonomous driving, robotic scene understanding, etc.

See our [CVPR'18 spotlight paper (acceptance rate<10%)], [code(in pycaffe)] .

KCNet overview

Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling

We propose two new operations, Kernel Correlation and Graph Pooling, to efficiently and robustly improve PointNet, useful for autonomous driving, robotic scene understanding, etc.

See our [CVPR'18 paper], [code(in pycaffe)] .

CASENet overview

CASENet: Deep Category-Aware Semantic Edge Detection

The state-of-the-art multi-label semantic boundary detection neural network, useful for autonomous driving, robotic scene understanding, etc.

See our [CVPR'17 paper], [code] .

DMT results

Direct Multichannel Tracking

A monocular VSLAM algorithm extending LSD-SLAM's input from single-channel gray image to multi-channel feature image.

See our [3DV'17 paper] .

DAL results

Deep Active Learning for Civil Infrastructure Defect Detection and Classification

A deep resisual convolutional neural network trained with our positive-based active learning strategy for multiple types of infrastructure defect detection, including cracks, deposit, and water leakage.

See our [IWCCE'17 paper], [slides] .

Contour Resampling results

Contour-Enhanced Resampling of 3D Point Clouds Via Graphs

An efficient point cloud resampling strategy using graph signal processing to reduce storage and computation cost for processing and visualizing large-scale 3D point clouds.

See our [ICASSP'17 paper] .

FasTFit overview

FasTFit: A fast T-spline fitting algorithm

The fastest T-spline fitting algorithm on CPU for organized point clouds, useful for reverse engineering, as-built BIM, point cloud compression, etc.

See our [JCAD'17 paper] .

MASfM Results

Marker-Assisted Structure from Motion for 3D Environment Modeling and Object Pose Estimation

A marker-assisted 3D reconstruction system modeled by camera-marker network, useful for multi-marker based pose estimation for AR/VR/Robotics/Camera Calibration/etc.

See our [CRC'16 paper] [code] .

PEAC results

Fast Plane Extraction in Organized Point Clouds Using Agglomerative Hierarchical Clustering

The fastest plane detection algorithm on single-core CPU (>35Hz for VGA size) for organized point clouds.

See our [ICRA'14 paper], [slides], [C++ codes] .

AMPE prototype

Vision-Based Articulated Machine Pose Estimation for Excavation Monitoring and Guidance

A prototype excavation monitoring system using markers and cameras for estimating the bucket position.

See our [ISARC'15 paper] .

Robotic Assembly Experiments

Towards Autonomous Robotic In-Situ Assembly Using Monocular Vision

A robotic prototype using AR markers to enable automatic in-situ assembly from an algorithmic architecture design with a single camera.

See our [ISARC'14 paper] (Best Paper Award) .

FollowMe system overview

Human-Robot Integration For Pose Estimation And Semi-Autonomous Navigation

A robotic prototype using AR markers to enable automatic dynamic target following with a single webcam.

See our [ISARC'13 paper] .

MARvigator system overview

MARvigator: Augmented Reality Markers as Spatial Indices

A new methodology for using Augmented Reality (AR) fiducial markers as spatial indices whose global positions and orientations are known in advance, presenting information in AR at a set of discrete critical spatial locations.

See our [CONVR'12 paper] .

PPSLAM results

Point-Plane SLAM for Hand-Held 3D Sensors

A robust feature-based RGBD-SLAM algorithm using both points and planes for robust camera pose estimation and 3D environment reconstruction.

See our [ICRA'13 paper] .

KEG tracker experiment result

KEG Plane Tracker for AEC Automation Applications

A single camera pose estimation algorithm for planar environment, combining two global constraints (geometric and appearance) to prevent tracking errors from propagating between consecutive frames, achieving high accuracy, stability, and robustness.

See our [CACAIE'13 paper] .

Semi-Auto SVR experiment result

Semi-Auto SVR: Semi-Automatic Single View Reconstruction

A semi-automatic reconstruction of buildings into piecewise planar 3D models from a single image with user input of line drawings, using J-linkage-based automatic vanishing point detection.

See our [CONVR'10 paper] .