Research Scientist @NVIDIAAI ; Working on Generative AI ๐ PhD @ Tel-Aviv University. Views are my own.
Israel
Joined May 2017
- Tweets112
- Following443
- Followers401
- Likes897
Pinned Tweet
๐ Happy to share our latest work: Bootstrap Your Generator: Unpaired Visual Editing with Flow Matching (accepted to #ICML2026)!
TL;DR: We train image and video editing models without any paired data.
No ground-truth edit data, no external reward models.
Yoad Tewel retweeted
Project page: research.nvidia.com/labs/parโฆ
Code and model: github.com/NVlabs/ProtoMotioโฆ
Co-led with Yifeng Jiang, alongside our fantastic collaborators @AlecSaysThings , @YiShi_333 , @UmarIqb , @GalChechik , and @xbpeng4 .
Yoad Tewel retweeted
if you want to come post-docing with me at the Technion, here's your chance: azrielifoundation.org/fellowโฆ
Always wanted to come work with me and my students but lacked funding? Youโre out of excuses ๐
The Azrieli Foundation (@azrielifdn) is sponsoring visiting PhD positions (up to 4 months) and postdocs (up to 3 years).
azrielifoundation.org/fellowโฆ
This is a very cool project led by @NoamRot !
Video models are the future of animation and synthetic data generation. They have a strong understanding of 3D structure and diverse motion.
We leverage their priors to co-generate video and 3D motion for any rig!
๐ฌ Excited to share our paper: Joints, Camera, Action! (JCA)
JCA generates video & 3D motion together for any rig!
Video models generate motion you can watch. But what if you could use that motion to animate a 3D asset, with motion that responds to its surroundings?
research.nvidia.com/labs/parโฆ
๐งต
Yoad Tewel retweeted
๐ฌ Excited to share our paper: Joints, Camera, Action! (JCA)
JCA generates video & 3D motion together for any rig!
Video models generate motion you can watch. But what if you could use that motion to animate a 3D asset, with motion that responds to its surroundings?
research.nvidia.com/labs/parโฆ
๐งต
Yoad Tewel retweeted
ืจืฆืืชื ืืืืืง ืขื ืืืืจื ืืขืืจืืช ืืคืื ืฉืืกืืกืืจ ืืงืืื ืืขืืจืชื ืืจืืฉืื ื ืืืื ืืื ืืช ืืงืืจืก ืฉืื Modern Computer Vision. ืืืงืืืืช ืฉื ืืืจืฆืืืช (ืืขืืจืืช) ืืื ืืืืืจ ืืืื ืื ืืืชืจ ืืงืืจืก. ๐
ืืฉืงืขืชื ืืืื ืืงืืจืก ืื ืืกืืชื ืืืฆืืจ ืืช ืืงืืจืก ืฉืืืืืื ืฉืืื ืื ืืชืืจ ืกืืืื ื. ืืงืืจืก ืืฉืื Deep learning ืืจืืืื (ืึพ70%) ืขื ืืกืืืืช ืฉื ืจืืืื ืงืืืกืืช ืืขืืืื ืืืชืืช (ืึพ30%). ืืชืืืืื ืืืฉ ืืืืคืืืงื ืืจื CV, ML ืงืืืกืืื (ืืืืื ืืื ืฉืืฉ ืืื ืืฉืืืืช ืืฉืืืืฉ ืืืืืจื ื) ืืขืืืื ืืืชืืช ืืืืืขืื ืืฉืืืืช ืืขืืื ืืืช, ืขื ืืืฉ ืขื ืืื ื ืืืืชืืช ืืขืืืง, ืืื ืืืืืฆืื ืืืชืืืืงื ืขื ืืื ืฉืคืืืช ื ืืคื ืืคื ืืืืื.
ืฉืืขืืจื ืืืืช (ืฉืืืื ืื ืื ืื ืืืืจืื) ืืืฉืงืขืื ืืื ืืืฉ ืืื ืคืืืืกืืคืื ืืฉืืื. ืื ืืฉืืื ืืืืืงืช ืืฉืืืฉื ืืืงืื:
Theory, from-scratch, applied.
ืื ืืืืืืจ ืืจืืฉืื ืฉื ืืงืืจืก ืื ืืืชืื ื ืืขืืืืช ืคื ืืฉื.
ืืืืืจื ืื ืืืืช ืฉืืจืื ืืชืจืื: ืืกืืกืืจ ืืืืื ืืงืืจืก ืืขืืืจ (ืื ืจืื) ืืื ืืืืช. ๐
Yoad Tewel retweeted
1/5 Excited to introduce IR4RL!
Image-to-code models build images step by step. We show that visual progress provides a powerful learning signal enabling RL post-training from intermediate renders.
With amazing @kate_feingold, @phillip_isola & @talidekel ir4rl.github.io/
Yoad Tewel retweeted
Meet DyRAD: Radar NVS for dynamic driving scenes.
๐ก๐ Radar directly senses motion through Doppler, but prior radar NVS methods ignore this signal where it matters most: dynamic scenes.
๐ dyrad-nvs.github.io
๐ arxiv.org/abs/2609.39841
๐ป github.com/Dyrad-NVS/DyRAD
๐งต
Yoad Tewel retweeted
I'm hiring a postdoc! The position is funded by my new ERC Starting Grant, GDLNeuralArt, and based in my group at the Technion.
We work on learning on weight spaces, internal representations and gradient spaces: treating neural networks themselves and their artifacts as data. ๐งต
Yoad Tewel retweeted
๐ฅณ Excited to share UniProbe: A learnable, token-level hallucination detection for large VLMs ๐
UniProbe looks inside a VLMโs internals to identify hallucinated response tokens and mitigate them before they reach the user.
๐ 55% fewer hallucinations
โก Just 1.06ร latency
I'm going to present ByG at #ICML26 today (Thursday) @ 2:30pm - poster #2615!
Stop by if you're interested in unsupervised editing with flow matching!
Project page: research.nvidia.com/labs/parโฆ
Details below โฌ๏ธ
๐ Happy to share our latest work: Bootstrap Your Generator: Unpaired Visual Editing with Flow Matching (accepted to #ICML2026)!
TL;DR: We train image and video editing models without any paired data.
No ground-truth edit data, no external reward models.
Yoad Tewel retweeted
I'm going to present Fast-AR at #ICML26 tomorrow (Tuesday) @ 2pm - poster #1204!
Come and say hi ๐
If you're around this week, feel free to DM me.
Project page: dvirsamuel.github.io/fast-auโฆ
Details below โฌ๏ธ
Yoad Tewel retweeted
"Spanning the Visual Analogy Space with a Weight Basis of LoRAs" is heading to #ECCV2026๐ฅณ
We learn to dynamically mix LoRAs to improve analogy-based image editing.
๐research.nvidia.com/labs/parโฆ
w/ the amazing @RinonGal @HaggaiMaron @GalChechik at @nvidia & @t_michaeli๐คฉ
@eccvconf
ALT Given a prompt and an image triplet a,a',b that visually describes a desired transformation, LoRWeB dynamically constructs a single LoRA from a learnable basis of LoRA modules and produces an editing result b' that applies the same analogy to the new image.
Yoad Tewel retweeted
This was fun work! A cool part of ByG is making cycle consistency work for denoising models. The missing piece was a Straight-Through trick. ST usually trains quantization or LLM sampling; here, ST lets reverse pass see a clean generated video while grads flow through a surrogate
๐ Happy to share our latest work: Bootstrap Your Generator: Unpaired Visual Editing with Flow Matching (accepted to #ICML2026)!
TL;DR: We train image and video editing models without any paired data.
No ground-truth edit data, no external reward models.
Thank you @_akhaliq for featuring our work!
teapawt.net/_akhaliq/status/206236โฆ
For more details on our bootstrapping mechanism, semantic loss, gradient routing, and other technical bits, check out the paper!
Project page: research.nvidia.com/labs/parโฆ
Paper: arxiv.org/abs/2606.03911