Prune Truong

I am a senior research scientist at Google, in the Semantic Perception team of Federico Tombari.

I received my PhD in 2024, from ETH Zurich in the Computer Vision Lab, supervised by Prof. Luc Van Gool and Dr. Martin Danelljan. My PhD was focused on Computer Vision and its applications, especially in the tasks of image matching, 3D reconstruction, pose estimation and novel-view rendering.

During my PhD I had the opportunity to intern both in Google working with Federico Tombari, and in Microsoft Mixed Reality & AI Labs, hosted by Marc Pollefeys. I am also one of the recipient of the Apple AI/ML PhD Fellowship of 2022.

Prior to that, I obtained a Master’s degree in Mechanical Engineering with honors at ETH Zurich. I also conducted an internship at RetinAI focused on computer vision applied to medical images.

Email  |  CV  |  GitHub  |  Google Scholar  |  LinkedIn  |  Twitter

headshot
News
My research
VIST3A: Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator
Hyojun Go, Dominik Narnhofer, Goutam Bhat, Prune Truong, Federico Tombari, Konrad Schindler
Arxiv 2025
citation | website | paper | code

Meet VIST3A — Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator, a new framework that connects the best of two worlds: 🎥 Video diffusion models for rich latent visual generation, and 🌍 Feed-forward 3D models (like VGGT, AnySplat, or MVDUSt3R) for geometric reconstruction.

AnyUp: Universal Feature Upsampling
Thomas Wimmer, Prune Truong, Marie-Julie Rakotosaona, Michael Oechsle, Federico Tombari, Bernt Schiele, Jan Eric Lenssen
Arxiv 2025
citation | website | paper | code

We introduce AnyUp, a method for feature upsampling that can be applied to any vision feature at any resolution, without encoder-specific training.

M2SVid: End-to-End Inpainting and Refinement for Monocular-to-Stereo Video Conversion
Nina Shvetsova, Goutam Bhat, Prune Truong, Hilde Kuehne, Federico Tombari
3DV 2026
citation | paper

We tackle the problem of monocular-to-stereo video conversion and propose a novel architecture for inpainting and refinement of the warped right view obtained by depth-based reprojection of the input left view.

One2Any: One-Reference 6D Pose Estimation for Any Object
Mengya Liu, Siyuan Li, Ajad Chhatkuli, Prune Truong, Luc Van Gool, Federico Tombari
CVPR 2025
citation | paper | code

We propose a novel method, One2Any, that estimates the relative 6-degrees of freedom (DOF) object pose using only a single reference-single query RGB-D image, without prior knowledge of its 3D model, multi-view data, or category constraints.

SPARF: Neural Radiance Fields from Sparse and Noisy Poses
Prune Truong, Marie-Julie Rakotosaona, Fabian Manhardt, Federico Tombari
CVPR 2023 - Highlight (top 2.5%)
citation | paper | project page | video (8 min) | teaser video | poster | code

We propose SPARF, a joint pose-NeRF refinement approach, applicable to extreme scenarios with only 2/3 input views and noisy camera poses. SPARF is the only method producing realistic novel-view synthesis from as few as 2 input images with noisy poses.

Refign: Align and Refine for Adaptation of Semantic Segmentation to Adverse Conditions
David Bruggemann, Christos Sakaridis, Prune Truong, Luc Van Gool
WACV 2023
citation | paper | code

We propose Refign, a generic extension to self-training-based domain adaptation methods for semantic segmentation.

PDC-Net+: Enhanced Probabilistic Dense Correspondence Network
Prune Truong, Martin Danelljan, Radu Timofte, Luc Van Gool
TPAMI 2023
citation | paper | project page | code

We propose an approach for estimating dense correspondences between two views along with a confidence map. We extend PDC-Net with new applications to image-based localization, 3D reconstructions and texture-transfer.

Probabilistic Warp Consistency for Weakly-Supervised Semantic Correspondences
Prune Truong, Martin Danelljan, Fisher Yu, Luc Van Gool
CVPR 2022
citation | paper | teaser video | poster | code

We propose a weakly-supervised training strategy for learning semantic correspondences. We introduce a weakly-supervised training objective applied to probabilistic mapping as well as an approach to model and identify occluded and unmatchable regions.

Warp Consistency for Unsupervised Learning of Dense Correspondences
Prune Truong, Martin Danelljan, Fisher Yu, Luc Van Gool
ICCV 2021 - Oral (top 3.0%)
citation | paper | teaser video | poster | code

We propose an unsupervised training objective for learning to regress the dense correspondences relating a pair of images. Our loss leverages real image pairs without invoking the photometric consistency assumption. Unlike previous approaches, it is capable of handling large appearance and view-point changes.

Learning Accurate Dense Correspondences and When to Trust Them
Prune Truong, Martin Danelljan, Luc Van Gool, Radu Timofte
CVPR 2021 - Oral (top 4.0%)
citation | paper | project page | teaser video | poster | slides | code

We develop a flexible probabilistic approach that jointly learns the dense correspondence prediction and its uncertainty. We parametrize the predictive distribution as a constrained mixture model and develop an architecture and training strategy tailored for robust and generalizable uncertainty prediction in the context of self-supervised training.

GOCor: Bringing Globally Optimized Correspondence Volumes into Your Neural Network
Prune Truong, Martin Danelljan, Luc Van Gool, Radu Timofte
NeurIPS 2020
citation | paper | teaser video | CV Talks video | code

We propose GOCOr, a fully differentiable dense matching module, acting as a direct replacement to the feature correlation layer. The correspondence volume generated by our module is the result of an internal optimization procedure that explicitly accounts for - and suppressed - similar regions in the scene.

GLU-Net: Global-Local Universal Network for dense flow and correspondences
Prune Truong, Martin Danelljan, Luc Van Gool, Radu Timofte
CVPR 2020 - Oral (top 5.7%)
citation | paper | teaser video | slides | poster | oral video | code

We propose GLU-Net, a unified architecture to estimate dense correspondences between any image pair, i.e. different views of the same scene, consecutive frames of a video or even different instances of the same object category.

GLAMpoints: Greedily Learned Accurate Match points
Prune Truong, Stefanos Apostolopoulos, Agata Mosinska, Samuel Stucky, Carlos Ciller, Sandro De Zanet
ICCV 2019
citation | paper | poster | code

We propose a training strategy for keypoint detection, applicable to low-quality and textureless images, frequent in the medical domain. We learn keypoint detection by training directly for the final matching accuracy instead of indirect metrics such as repeatability.

Invited Talks
  • 2023: Dense Matching and Its Applications, Invited talk for the Swedish WASP program, Zurich.

  • 2022: Dense Matching, Invited talk at Google, Semantic Perception, Zurich.

  • 2022: Dense Matching, Invited talk at Microsoft, Mixed Reality and AI Lab, Zurich.

  • 2021: PDC-Net and matching challenge, Image Matching Workshop: Local Features & Beyond, CVPR 2021, Virtual.

  • 2021: PDC-Net (CVPR 2021), Reading group of Dr. Krystian Mikolajczyk at Matchlab, Virtual.

  • 2021: GOCor (NeurIPS 2020), Computer Vision Talks, Virtual.

template adapted from this awesome website