Slava Elizarov

362 posts

Slava Elizarov

Slava Elizarov

@DoctorDukeGonzo

Staff Research Scientist @canva, ex-Unity | Generative models, Computer Graphics

Germany Katılım Nisan 2012
4.7K Takip Edilen821 Takipçiler
Sabitlenmiş Tweet
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
Does 3D generation always have to be either slow or complex and data-hungry?🤔 We don’t think so! With Geometry Image Diffusion, we’re all about reusing (and recycling ♻️) what already works — making it faster and easier by reducing complexity and data needs 🚀(1/10)
English
20
61
451
67.5K
Slava Elizarov retweetledi
Ethan
Ethan@torchcompiled·
Have you ever gotten tired of boring plain linear layers and wanted a more complex function? We find that attaching low rank nonlinear residual functions can significantly accelerate pretraining, with an identified variant, CosNet, consistently observing 20+% wallclock speedup!
Ethan tweet media
English
16
47
260
38.8K
Nikita Karaev
Nikita Karaev@n_karaev·
Check out our latest work, VGGT! ⚡️ It’s a fast transformer that predicts cameras, point maps, depth and 3D point tracks. VGGT + CoTracker significantly outperforms CoTracker in 2D point tracking.
Jianyuan@jianyuan_wang

Introducing VGGT (CVPR'25), a feedforward Transformer that directly infers all key 3D attributes from one, a few, or hundreds of images, in seconds! No expensive optimization needed, yet delivers SOTA results for: ✅ Camera Pose Estimation ✅ Multi-view Depth Estimation ✅ Dense Point Cloud Reconstruction ✅ Point Tracking Project Page: vgg-t.github.io Code & Weights: github.com/facebookresear…

English
1
9
57
14.5K
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
I’m currently exploring new job opportunities🧑‍🔬 My work revolves around text-to-3D with Geometry Images, generative UV mapping, multi-view models for texturing, and other genAI applications in graphics. I’d love to discuss how I can contribute to your research efforts!
Slava Elizarov@DoctorDukeGonzo

Does 3D generation always have to be either slow or complex and data-hungry?🤔 We don’t think so! With Geometry Image Diffusion, we’re all about reusing (and recycling ♻️) what already works — making it faster and easier by reducing complexity and data needs 🚀(1/10)

English
4
10
46
9.8K
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
@gadelha_m Thank you! That's an excellent question. We rescaled each 2D chart with respect to the area of the corresponding surface, making the mapping close to an equal-area projection (engineering.purdue.edu/cdesign/wp/dee…), It helps a lot. As for the alignment of UV charts, we don't use any.
English
0
0
0
82
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
Does 3D generation always have to be either slow or complex and data-hungry?🤔 We don’t think so! With Geometry Image Diffusion, we’re all about reusing (and recycling ♻️) what already works — making it faster and easier by reducing complexity and data needs 🚀(1/10)
English
20
61
451
67.5K
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
@Ronnie36925798 Our model operates at 768x768 vs 64x64 for Omages (btw, they have cool boundary snapping trick) We used 100k subset of Objaverse vs ABO dataset (8k samples) Our method can generate highly novel 2D layouts, going beyond what’s present in the training dataset (3/3)
English
0
1
1
184
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
@Ronnie36925798 While the authors explored class-conditioned generation, our focus is on Text-to-3D. We also use Collaborative Control to leverage the rich 2D prior of Stable Diffusion instead of training from scratch (2/3)
English
1
1
1
194
Slava Elizarov
Slava Elizarov@DoctorDukeGonzo·
P.P.S. We recommend you check out Omages (omages.github.io) by @yan_xg , an awesome concurrent work that also explores geometry images (called "Omages") for 3D generation. We believe GIMs have a bright future in deep learning — let’s bring it forward together 🚀
English
1
1
10
1.6K

Keşfet

@n_karaev @cortwave @yan_xg @CiaraRowles1 @esx2ve @danteCIM @bostadynamics @xdralex