RESEARCH / PUBLICATIONS
How can machines understand physical space and create coherent visual worlds? Explore my research in generative AI and geometric vision through plain-language summaries, method explanations and links to the original work.
25 papers, preprints, a thesis and a technical report · Notes updated September 26, 2026.
View my Google Scholar profile ↗
25 research records
From sketches and stories to editable environments, concept art and expressive digital humans.
ECCV 2026, 130–147 · 2026
StoryBlender turns a story into editable 3D storyboards while keeping characters, assets and spatial relationships consistent across shots. A persistent scene representation connects narrative planning with engine-verified scene construction.
arXiv:2602.10516 · Preprint · 2026
3DXTalker combines identity, audio-driven lip motion, emotion and head-pose control in expressive 3D talking avatars. Its representation separates these factors so that speech animation and semantic control can work together.
Transactions on Machine Learning Research · 2026
T³-S2S turns sketches into controllable multi-object scene concept art through training-free interventions in diffusion attention and sketch conditioning. It also supports paired isometric and terrain-oriented views.
arXiv:2408.04567 · Preprint · 2024
Sketch2Scene converts casual sketches into interactive 3D game scenes by combining diffusion-based concept generation, scene interpretation and procedural construction in a game engine.
Memory, action control and visual continuity across camera movements and multiple clips.
NeurIPS 2026 · 2026
Robust Dreamer improves long autoregressive video rollouts with action control by storing diffusion latents in a 3D Gaussian memory and training the generator to handle errors in its own historical memory.
arXiv:2603.23413 · Preprint · 2026
I3DM preserves scene consistency in generated video by retrieving useful historical views and injecting spatially aligned memory features, without requiring an explicit 3D reconstruction of the scene.
arXiv:2510.21696 · Preprint · 2025
BachVid generates multiple video clips with consistent backgrounds and characters using a training-free reuse mechanism. It carries identity-related evidence from an initial video into later clips while allowing actions to change.
Multi-camera calibration, visual-inertial mapping, geometric reconstruction and vehicle motion.
IEEE Robotics and Automation Letters · 2025
ColonAdapter adapts a visual geometry foundation model to colonoscopy, using self-supervised geometric and photometric constraints to estimate depth, camera pose and dense 3D structure.
arXiv:2503.03230 · Preprint · 2025
OpenGV 2.0 uses vehicle motion priors to improve surround-view camera calibration, relative pose estimation and SLAM, including configurations with little overlap between cameras.
ICRA 2024 · 2024
MAVIS combines multiple cameras with an IMU for visual-inertial SLAM. Its exact SE₂(3)-based IMU pre-integration strengthens motion estimation during fast rotations and longer integration intervals.
Technical report · Hilti SLAM Challenge 2023 · 2023
TXR-SLAM is the XR Penguin team's vision/IMU-only entry to the Hilti SLAM Challenge 2023, combining multi-camera tracking, sliding-window optimization, loop closure and full trajectory refinement.
2022 IEEE International conference on robotics and automation (ICRA) · 2022
This paper generalizes the hand–eye calibration constraint to multi-perspective camera systems, providing a closed-form route to calibrating camera rigs against an externally tracked reference.
2021 IEEE International conference on robotics and automation (ICRA) · 2021
This paper uses B-splines to incorporate smooth, non-holonomic vehicle motion into visual localization and mapping, refining trajectories within six-degree-of-freedom bundle adjustment.
Doctoral thesis · Australian National University · PQDT-Global · 2021
This PhD thesis studies localization and mapping with surround-view multi-camera systems and event cameras, linking sensor geometry, vehicle motion constraints and continuous-time estimation.
2020 IEEE International conference on robotics and automation (ICRA) · 2020
This paper improves frame-to-frame motion estimation for vehicle-mounted surround-view cameras, addressing planar-motion degeneracies with a geometric two-view optimization.
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition · 2019
This paper estimates non-holonomic ground-vehicle motion from one tracked feature across multiple views, using a constrained motion model to derive a compact polynomial solution.
International Conference on Computer Vision Systems · 2017
This paper studies metric-scale initialization for visual odometry with non-overlapping multi-perspective cameras, including the motion degeneracies that can make scale unobservable.
Asynchronous sensing for motion estimation, tracking, place recognition and frame interpolation.
IEEE Transactions on Robotics, 40, 1600–1616 · 2024
This work tracks an event camera in six degrees of freedom against a map built using another sensing modality, combining semidense geometric alignment with polarity-aware event representations.
IROS 2023 · 2023
This paper combines RGB-guided motion estimation with event-based refinement for video frame interpolation, using incremental intermediate-frame synthesis to handle larger temporal gaps.
arXiv:2307.01047 · Preprint · 2023
This work recognizes places from event-camera observations by retrieving matches in a conventional image database, connecting event sensing with existing visual reference maps.
Sensors · 2022
This event-camera visual odometry method jointly estimates motion and scene structure by warping viewing rays continuously and maximizing their concentration in a 3D volume.
2022 IEEE International conference on robotics and automation (ICRA) · 2022
DEVO combines a depth camera with an event camera for visual odometry, aligning semidense geometry with event-derived time surfaces to track motion in challenging lighting and fast movement.
2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · 2021
Dynamic Event Camera Calibration uses motion between an event camera and a conventional calibration pattern to estimate camera parameters, followed by continuous-time refinement.
IEEE Transactions on Pattern Analysis and Machine Intelligence · 2021
This paper develops globally optimal contrast maximization for event-camera motion models, using branch-and-bound search with recursive bounds to avoid dependence on a favorable initial guess.
European Conference on Computer Vision · 2020
This work derives a globally optimal contrast-maximization solution for event-camera motion estimation, focusing on the fronto-parallel geometry of a downward-looking ground-vehicle camera.
No matching research. Try another topic, title or year.
* Equal contribution · # Corresponding author