Julian Chibane

Check out how to better understand and generate human motion sequences in our new work UniMotion @3DVconf '25!

It was a pleasure to give a talk at @Bosch_AI with @anna_khoreva and colleagues! Hope to see you in person soon.

We lift images to 2.5D, by predicting front and back depth. Then, an IF-Net [Chibane et al. CVPR'20] creates the final 3D reconstruction. Our key insight is that the depth estimation learns local, class agnostic reconstruction, allowing for generalization to novel classes.

Thrilled to receive the best paper honorable mention at 3DV'22 - together with Yongqin & @bharat_b7. Thanks to supervisors @zeynepakata, Bernt & @GerardPonsMoll1! Our method improves single image 3D reconstruction and generalizes to novel object classes. virtualhumans.mpi-inf.mpg.de/gin/

Box2Mask unlocks a large body of 3D detection datasets to be viable for learning instance segmentation. Checkout our paper, video and interactive visualization on our project page: virtualhumans.mpi-inf.mpg.de/box2mask/! (5/5)

For each point in the input scene, Box2Mask predicts the instance, parameterized as a bounding box - solely trained with coarse bounding box annotations. Votes are clustered via a novel algorithm called Non-Maximum Clustering, tailored to the bounding box representation.

Check out our new work: 𝐁𝐨𝐱𝟐𝐌𝐚𝐬𝐤 virtualhumans.mpi-inf.mpg.de/box2mask/ We’re asking the question: Are 3D bounding box (BB) annotations enough to train dense 3D semantic instance segmentation? We find: Yes! (1/5)

From our group, Bharat Lal Bhatnagar, Garvita Tiwari, Ilya Petrov, Riccardo Marin, and I are Outstanding Reviewers at CVPR22! cvpr2022.thecvf.com/outstanding-re… Also, many other colleagues are acknowledged! Congrats everyone!

@davidstutz92 @DeepMind Congrats David! Very cool!

Now that my visa is approved, I am excited to share that I will join @DeepMind as research scientist in the robust and verified AI team in April. Looking forward to this next chapter!

Thrilled to receive the Facebook/@Meta Research 𝗙𝗲𝗹𝗹𝗼𝘄𝘀𝗵𝗶𝗽 𝗔𝘄𝗮𝗿𝗱 for my work on capture and synthesis of 3D scenes and humans. Extremely grateful for the support of my advisor @GerardPonsMoll1 and Bernt Schiele. Congrats to all fellows! research.facebook.com/fellows/chiban…

Our IF-Networks for 3D shape and texture reconstruction won again #1 place at SHARP'21 #CVPR21 - together with @GerardPonsMoll1 In our experience IF-Net is a powerful 3D processing tool - check it at virtualhumans.mpi-inf.mpg.de/ifnets/ Thanks @AouadaDjamila @Artec3DScanners + organizers!

Join us in 20 minutes at #CVPR2021 and find out about Stereo Radiance Fields (SRF) How to make NeRF - generalise to new scenes? - work with only 10 views? - and integrate classical stereo to NeRF? Code/Paper: virtualhumans.mpi-inf.mpg.de/srf/