Chen Zhao

🎥 Check out the presentation video below for our paper "Self-Ensembling Gaussian Splatting for Few-Shot Novel View Synthesis", which has been accepted as an Oral at ICCV 2025! 🚀 🔗 Youtube: youtube.com/watch?v=UOYMtR… #ICCV2025 #3DV #computervision #machinelearning

Two papers have been accepted by CVPR 2024 🎉 DVMNet: Computing Relative Pose for Unseen Objects Beyond Hypotheses arxiv.org/pdf/2403.13683… HOISDF: Constraining 3D Hand-Object Pose Estimation with Global Signed Distance Fields arxiv.org/pdf/2402.17062… #CVPR2024 #computervision

One paper was accepted by ICLR2024. We handle the problem of unseen object pose estimation. Our method can estimate the object pose in the query image, given a single reference. It achieves SOTA performance on several datasets. Project page:sailor-z.github.io/projects/ICLR2… #ICLR2024

Pascal Fua has openings for #PhD students in his group. Find out more about @FuaPv's #research epfl.ch/labs/cvlab/ and our world-leading @EPFL #EDIC #computerscience PhD program: go.epfl.ch/phd-edic

Amazing experience in Israel. Thanks @eccvconf for all the events. Got lots of new friends during the conference. Hope to see you again in the next one.

@pesarlin @PhilippCSE @mapo1 Congratulations!

🚨 This week at #ICCV2021, check out "Pixel-Perfect Structure-from-Motion with Featuremetric Refinement" w/ @PhilippCSE V. Larsson @mapo1 ➡️ oral & *best student paper award* Website: psarlin.com/pixsfm Paper: arxiv.org/abs/2108.08291 Video: youtu.be/2HuCMuraFk0 thread ⬇️

Progressive Correspondence Pruning by Consensus Learning youtu.be/tPEkLDSE_6g via @YouTube

Amazing talk by Serge Belongie

One paper about correspondence pruning has been accepted by #iccv2021! Project page: sailor-z.github.io/projects/CLNet…

We are looking for post-doctoral fellows in several areas of computer vision. Please send your CV with a brief motivation letter to ariane.staudenmann@epfl.ch. #DeepLearning #ComputerVision epfl.ch/labs/cvlab/

@quantombone By the way, if I understand correctly, the reference image is obtained by image retrieval like NetVlad, the reference image should be highly overlapped with the query ideally. Is it an over-kill to use a complicated network to refine the pose?

Back to the Feature: Learning Robust Camera Localization from Pixels to Pose This paper is based on the alignment of deep multiscale features and combines the best of deep learning with multiview geometry. Checkout those learned heatmaps! arxiv.org/abs/2103.09213 #computervision

@quantombone Thanks for releasing the amazing work. I got some questions. Does the setup make sense in practice, assuming that the 3d model is known?

@kwangmoo_yi My feeling was like, reviewers and ac did not care about my rebuttal at all. I was wondering if they read it or not, because I'm frustrated by getting the final justification with only one sentence.

So.... the feeling when you see reviewers' final justification based on things that are already rebutted in the main paper/rebuttal... and in BOLD. Maybe I should do CAPS as well next time.

My open letter to all fans facebook.com/10000158330719…