Research Scientist · CV / 3D / Generative AI

Building perceptual systems for the physical world.

I am a Senior Researcher at Qualcomm and I got my PhD in Computer Science from the University of Oulu, Finland. At Qualcomm AI Research Viet Nam, I am currently leading a team working on World Model and Robot Learning.

Email CV Scholar GitHub

Research focus

My work spans 3D reconstruction, neural scene representations, depth estimation, and controllable image generation. I am especially interested in systems that are both physically grounded and computationally efficient.

Current themes

Generalizable novel view synthesis
3D content generation and editing
Geometry-aware image and video modeling
Efficient training and inference for vision systems

Selected work

Research portfolio

ECCV 2026

Cross-Space Distillation

Teaching one-step students with modern diffusion teachers using a latent-space bridge for cross-space distillation.

ArXiv BibTeX

ECCV 2026

PixGS

Direct 3D Gaussian splat generation with pixel-space diffusion for high-quality, single-stage 3D asset synthesis.

ArXiv BibTeX

CVPR 2026 Highlight

SwiftTailor

Efficient 3D garment generation with geometry-image representation and pattern-aware synthesis.

Project ArXiv

CVPR 2026 Highlight

PixelRush

Training-free, ultra-fast high-resolution image generation with a one-step diffusion pipeline.

Project ArXiv

CVPR 2026

Ar2Can

Multi-human generation with a spatial planner and identity-preserving diffusion renderer.

ArXiv BibTeX

In submission

ModeDreamer

Guiding score distillation with reference-image prompts for more stable and diverse text-to-3D generation.

ArXiv BibTeX

CVPR 2025

SharpDepth

Sharpening metric depth predictions using diffusion distillation for accurate and crisp geometry.

ArXiv BibTeX

AAAI 2025

Semi-supervised 3D Semantic Scene Completion

Using 2D vision foundation models to reduce the need for costly voxel-wise labels in 3D occupancy prediction.

ArXiv BibTeX

ECCV 2024

DiverseDream

Diversifying text-to-3D synthesis by augmenting prompts and modeling joint generation.

Project ArXiv

TPAMI 2024

CG-NeRF

Generalizable neural radiance fields for fast novel-view synthesis with a coarse-to-fine renderer.

ArXiv BibTeX

PhD Thesis 2023

Neural Scene Representations for Learning-Based View Synthesis

Introducing neural scene representations for efficient and compact learning-based novel view synthesis.

Link

ECCV 2022

Free-Viewpoint RGB-D Human Performance Capture

Learning dense features in novel views and inpainting occluded regions for photorealistic human rendering.

Project BibTeX

3DV 2021

RGBD-Net

Predicting color and depth images for novel-view synthesis through a depth-aware generator.

ArXiv BibTeX

WACV 2021

LiDNAS

Efficient neural architecture search for lightweight monocular depth estimation.

ArXiv BibTeX

3DV 2021

Monocular Depth with Saliency and Hessian Loss

Improving monocular depth estimation with salient point priors and a Hessian-based loss.

ArXiv BibTeX

ICCV 2021

Boosting Monocular Depth with 3D Point Fusion

Guiding depth networks with extremely sparse point clouds for more robust geometry estimation.

ArXiv BibTeX

ACCV 2020

Sequential View Synthesis with Transformer

Using multi-view attention and a sequential decoder for consistent scene synthesis without retraining.

Paper BibTeX

ECCV 2020

Depth-Attention Volume

Using a non-local coplanarity prior to guide monocular depth estimation with planar structure awareness.

ArXiv BibTeX

SCIA 2019

Predicting Novel Views Using Generative Adversarial Query Network

Combining GQN with adversarial training for generative novel view synthesis.

ArXiv BibTeX

Sensors 2018

LightDenseYOLO

A fast and accurate marker tracker for autonomous UAV landing using visible-light camera input.

Paper BibTeX

Sensors 2017

Remote Marker-Based Tracking for UAV Landing

Reliable visible-light marker-based tracking for drone landing without GPS guidance.

Paper BibTeX