| CARVIEW |
Hello, I'm Tuan Duc Ngo
I am a third-year PhD student in Computer Science of UMass Amherst, USA, advised by Prof. Evangelos Kalogerakis and Prof. Chuang Gan. I was a research intern at Adobe Research. Previously, I interned at Snap Research, working with Dr. Chaoyang Wang on 4D reconstruction. Prior to my Ph.D., I was an AI Research Resident at VinAI Research, working closely with Dr. Khoi Nguyen. I received my B.E. degree in Computer Engineering from the Ho Chi Minh City University of Technology, Vietnam. Email: ductuan.ngo99 (at) gmail (dot) com
News
- May 2025: I will join Adobe Research as a Research Intern this summer.
- Feb 2025: 4Real-Video is accepted (highlight) to CVPR 2025.
- Jan 2025: DELTA is accepted to ICLR 2025, and the code is also released (Github)
- May 2024: I joined Snap Research as a Research Intern.
- Feb 2024: Open3DIS is accepted to CVPR 2024. We have also released the code.
- Jul 2023: GaPro is accepted to ICCV 2023. We have also released the code.
- Feb 2023: ISBNet is accepted to CVPR 2023. We have also released the code.
- Jul 2022: GeoFormer is accepted to ECCV 2022. We have also released the code.
Publications
4Real-Video: Learning Generalizable Photo-Realistic 4D Video Diffusion
A diffusion model that generates 4D video -- a grid of video frames with both time and viewpoint axes.
DELTA: Dense Efficient Long-range 3D Tracking for Any video
Capture the dense, long-range, 3D point trajectories from casual videos in a feed-forward manner
Open3DIS: Open-vocabulary 3D Instance Segmentation with 2D Mask Guidance
Tackle the open-vocabulary 3D point cloud instance segmentation by using 2D prior
GaPro: Box-Supervised 3D Point Cloud Instance Segmentation Using Gaussian Processes as Pseudo Labelers
Tackle the box-supervised 3D point cloud instance segmentation by using Gaussian Processes to generate pseudo labels
ISBNet: a 3D Point Cloud Instance Segmentation Network with Instance-aware Sampling and Box-aware Dynamic Convolution
Introduce an efficient sampling strategy and propose leveraging the bounding box as a geometric cue for the 3D point cloud instance segmentation
Geodesic-Former: A Geodesic-Guided Few-Shot 3D Point Cloud Instance Segmenter
Propose a new task, Few-shot 3D point cloud instance segmentation, and introduce a geodesic-based 3D instance segmenter
GAC3D: improving monocular 3D object detection with ground-guide model and adaptive convolution
Propose a new monocular 3D detection framework leveraging the ground plane model and depth-adaptive convolution
