CVMP

Special mention to Sophie Maxwell from CVSSP @peoplecentredai.bsky.social for the exceptional work she's put on organising this conference!

Also to our fantastic conference chairs Hansung Kim
@unisouthampton.bsky.social and Claudio Guarnera
@york.ac.uk

From mind-bending NeRFs and Gaussian Splats to the future of 3D meshes and virtual humans, the discussions were top-notch. Congratulations again to all our award winners! 🏆

It was fantastic to see industry and academia connect. We can't wait to see you all next year!

that 's all folks written on a red and blue background ALT: that 's all folks written on a red and blue background

And that's a wrap on #CVMP2025! 🎬 What an incredible few days at the BFI Southbank. A huge thank you to all our attendees, our four amazing keynote speakers, all the authors who presented their groundbreaking research, and our fantastic sponsors and volunteers.

Special mention to the runner ups Stephane Belemkoabga and Umar Farooq, it was a hard decision to chose a winner!

Congratulations to Paul Hinzer and team from @fraunhoferhhi.bsky.social for winning best paper award! 🏆🎉

Congratulations to Oliver Camilleri from CVSSP @peoplecentredai.bsky.social @uniofsurrey.bsky.social for winning the best short paper award! 🏆

The elephant in the room of 3D reconstruction is light! We can use diffusion models to tackle this, but they come with a range of problems, e.g., they are very compute intensive and expensive, there's identity loss and don't really have 3D consistency at the moment.

He argues for doing demo-centred research which necessitates going the last mile of research. For example, for NeRF going the last mile was tackling the robustness, scale, and speed.

Our last keynote of the conference #CVMP2025 is by Peter Hedman on The Last Mile of Research for Production-Ready View-Synthesis focusing on his work at Google and the most recent developments of NeRFs and Sora.

Next up, Philip Coulam-Jones from Disguise shows us the efforts they've done to bring performers into the virtual world and the challenges they've faced and overcome in "Evaluation of Depth-Based Volumetric Capture Methods in Production Environments"

For our Industrial session we have Elliott Round, CoFounder and CTO of M-XR, a 3D production technology company based in London. He talks about the challenges of capturing and representing different materials for production at a scale.
#CVMP2025

It's a really interesting and important area of research but there are still a lot of challenges and bottlenecks, with lots of future research avenues!

Creating 3D meshes is difficult and time consuming, but they are a good compact representation

Angela's research lab has been looking at how to apply the revolutionising technology of transformers to the generation of 3D meshes by creating a vocabulary of geometry

The first keynote today at #CVMP2025 is by Angela Dai @adai.bsky.social from 3DAI Lab at TU Munich

"Can Transformers speak geometry?"

There are lots of different 3D representations we use in learning but industry widely uses meshes.

And finally Paul Hinzer from @fraunhoferhhi.bsky.social presents "3D-LaSR: 3D-Aware Latent-Space Reenactment: Combining Expression Transfer and Semantic Editing" which uses a joint GAN-inversion technique to reenact 3D videos of avatars

Now we have Tengjiao Sun from Southampton University @unisouthampton.bsky.social presenting "Hi-RQCT: Hierarchical Residual-Quantized Causal Transformer for High-Quality 3D Human Motion Generation" which tackles real-time generation of long sequences with a 3 stage method

Next up we have Armin Mustafa presenting the work of Ayushi Dutta from CVSSP @peoplecentredai.bsky.social on "Realistic Clothed Human and Object Joint Reconstruction from a Single Image" where they also propose a new synthetic dataset of human-object meshes which their method uses for joint training

Our 3rd paper session of #CVMP2025 is on Virtual Humans

First up we have Guillaume Loranchet (InterDigital & Inria) showing how they tackle the creation of realistic avatars in motion with controllable identity in their work "Neural Implicit Avatar Conditioned on Human Pose, Identity and Gender"

Welcome to the second day of #CVMP2025

First day Vs secon day

Don't be put off by the rain, that's just London for you 🙄☔

Congratulations to all the winners 🏆🎉 👏🏽👏🏽👏🏽

The Collaboration Award goes to the European Media and Immersion Lab (EMIL). With Juhani Tenhunen from Aalta University, Volker Helzle from Filmakademie Baden-Wurttemberg, Narcis Pares from Universitat Pompeu Fabra @upf.edu , and Christof Lutteroth from University of Bath @uniofbath.bsky.social

Time for the #CVMP2025 technical awards!! 🏆

The Research Impact Award goes to Wieland Morgenstern, Florian Barthel, Anna Hilsmann, and Peter Eisert at @fraunhoferhhi.bsky.social & Humboldt University for their work on "Compact 3D scene representation via self-organising Gaussian grids"

From multi-layer images to multi-sphere images to 360 3D images to neural 3D video. From NeRF to Gaussian Splatting to scene reconstruction from LiDAR or shadows (PlatoNeRF) 😮

From omnidirectional stereo cameras to manifold cameras to 360 cameras to the "Eyeful Tower" 👀🎥

For the second keynote of #CVMP2025 "Unleashing immersive spaces by capturing, reconstructing, and rendering reality", Christian Richardt @cr333.bsky.social from Meta takes us in an exploration of immersive experiences through history to his work through the years

Finally, Athanasios Charisoudis from EPFL École Polytechnique Fédérale de Lausanne looks at capture and reconstruction from a point of view of usability and practicability with "A Fast Volumetric Capture and Reconstruction Pipeline for Dynamic Point Clouds and Gaussian Splats"

#EPFL #HSLU #CVMP2025

Now, Alp Orgun also from CVSSP at University of Surrey @peoplecentredai.bsky.social introduces his work on 3DGS relighting with "Dual Spherical Harmonics for 3D Gaussian Splatting: Novel View Synthesis with Dynamic Lighting"

Next up, Umar Farooq from CVSSP at University of Surrey @peoplecentredai.bsky.social presents his work on "Optimized 3D Gaussian Splatting using Coarse-to-Fine Image Frequency Modulation" to optimise Gaussian Splatting memory consumption

Our second session of the day is all about those Gaussian Splats

First, Stephane Belemkoabga from Inria presents "GS-Morph: Dynamic Novel View Synthesis via UDF-ARAP Gaussian Splat Morphing" tackling the incredibly ill-posed problem of monocular video reconstruction

#CVMP2025 #GaussianSplatting

Alexandros Neophytou comes from our sponsor @microsoft.com to show us their AI Cocreator integrated in Paint on Copilot+ PCs 🎨🖌️🖥️
#CVMP2025