Sadegh Aliakbarian

538 posts

Sadegh Aliakbarian banner
Sadegh Aliakbarian

Sadegh Aliakbarian

@aa_sadegh

Scientist @Microsoft

Cambridge, England Katılım Nisan 2015
185 Takip Edilen421 Takipçiler
Hanbyul (Han) Joo
Hanbyul (Han) Joo@jhugestar·
We present DAViD at #ICCV2025 . DAVid presents the first approach to learn 4D human-object interaction from synthetic videos, aligned with our lab’s recent efforts to leverage synthetic data for HOI understanding. Project Page: snuvclab.github.io/david/ Code is available.
Hyeonwoo Kim@Hyeonwoo__Kim

Excited to present our work “DAViD” at #ICCV2025! DAViD is a generative 4D human-object interaction model, which can generate novel HOI motions for various 3D objects, including multi-object interactions. (1/4)

English
2
3
24
2K
Artsiom Sanakoyeu
Artsiom Sanakoyeu@artsiom_s·
Mark @finkd is talking about our Imagine Flash right here. I won't lie, it feels really good when your CEO speaks about your work this way 🙂
English
2
1
38
3.8K
Sadegh Aliakbarian retweetledi
Amir Habibian
Amir Habibian@amir_habibian·
Five fellowships (each 40,000$) available for PhD students in European universities. More info: qualcomm.com/research/unive… DM if have any question.
Amir Habibian tweet media
English
2
11
81
12.3K
Sadegh Aliakbarian
Sadegh Aliakbarian@aa_sadegh·
Checkout "3DiFACE", our new generative model for Speech-driven 3D facial animation and editing 👇
English
0
0
10
767
Jiaxi Jiang
Jiaxi Jiang@cs_jiaxi_jiang·
@aa_sadegh @fatemeh_saleh @dopomoc (2) What is the form of the SE(3) loss? Is it a L2 loss for all elements in transformation matrix, geodesic loss, or 6D rotation representation? How much can it improve compared to previous work (local 6D rotation representation + FK position loss)?
English
2
0
0
67
Sadegh Aliakbarian
Sadegh Aliakbarian@aa_sadegh·
@cs_jiaxi_jiang @fatemeh_saleh @dopomoc That's a good question, while I don't have enough experiment to back it up, but computing the additional loss on global joint transformation matrices typically boosts the performance a lot. Euc distance for position and geodesic loos for rot mat (in 6D) is very effective
English
0
0
1
53
Sadegh Aliakbarian
Sadegh Aliakbarian@aa_sadegh·
@cs_jiaxi_jiang @fatemeh_saleh @dopomoc 2/n in addition to that (2) our SE(3) loss is contributing a lot when it comes to MPJPE. See fig (9) that removing this loss term leads to a model with MPJPE of >5cm. Finally (3) optimization on top of NeMo's output (table 6/7) yields further improvement.
English
1
0
1
81
Mohsen Ghafoorian
Mohsen Ghafoorian@mohsen_gh87·
@jujihong @ukrdailo We will also be giving demo's of some of our most recent XR perception developments at our Qualcomm booth 27B. Come visit us!
English
1
0
3
238
Mohsen Ghafoorian
Mohsen Ghafoorian@mohsen_gh87·
Excited to be heading to ICCV in Paris! Will be co-presenting two papers from my Qualcomm XRLabs team on improved depth perception and 3D reconstruction. * DG-Recon: Depth-Guided Neural 3D Scene Reconstruction
English
1
0
10
490
Sadegh Aliakbarian
Sadegh Aliakbarian@aa_sadegh·
Check out Imitator, a novel approach to generating personalised speech-driven facial animations. Come to our #ICCV2023 poster on Friday (Room "Nord" - 058) 😊
Justus Thies@JustusThies

Next week @ICCVConference, @balathambiraja will present his work Imitator which learns personalized speech-driven 3D facial animation. Web : balamuruganthambiraja.github.io/Imitator/ Code: github.com/bala1144/Imita… In collaboration with Ikhsanul Habibie, @aa_sadegh, @dopomoc, Christian Theobalt

English
1
6
36
5.9K
Sadegh Aliakbarian retweetledi
Peyman Milanfar
Peyman Milanfar@docmilanfar·
This ad is amazing - it challenges your assumptions and is technically impressive. Watch to the end. youtu.be/QVNZRHIZVL8
YouTube video
YouTube
English
4
24
106
26.7K

Keşfet

@jhugestar @artsiom_s @finkd @cs_jiaxi_jiang @fatemeh_saleh @dopomoc @jujihong @ukrdailo