Daniel Gilo

[Daniel Gilo @danielgilo1]

Conditional diffusion/flow models often produce outputs inconsistent with the very signal conditioning them. The error is easily measurable, yet models are never trained to act on it.

In FlowBender (now on arXiv), we train the model to correct its own errors. 🧵

[Or Litany @orlitany]

Meet DyRAD: Radar NVS for dynamic driving scenes.

📡🚗 Radar directly senses motion through Doppler, but prior radar NVS methods ignore this signal where it matters most: dynamic scenes.

🌐 https://t.co/a2iogtcQtw
📄 https://t.co/j15kt3GzGV
💻 https://t.co/AoWWeLTheX
🧵

[Or Litany @orlitany]

If you think differentiable rendering is all you need to animate your 3DGS avatars/assets, think again!

At ECCV? We invite you to come re-think together with us at poster #135 happening today at 10:30

#ECCV2026 #3DGS

[Daniel Gilo @danielgilo1]

Update: FlowBender code is public 🚀
✅ Depth → RGB
✅ Edge map → RGB
✅ Super-resolution
✅ JPEG restoration
✅ 3D texturing

Try it: https://t.co/BQnCarBqKB Project page: https://t.co/FnFWJ3Sr7y

[Daniel Gilo @danielgilo1]

Conditional diffusion/flow models often produce outputs inconsistent with the very signal conditioning them. The error is easily measurable, yet models are never trained to act on it.

In FlowBender (now on arXiv), we train the model to correct its own errors. 🧵

[Guy Bar-Shalom @GuyBarSh]

New blogpost out 📃

"Detecting LLM Misbehaviors from the Inside Out with Deep Learning on Structured Data" (https://t.co/tdcTr3waPZ) [1/8]

[Daniel Gilo @danielgilo1]

Thanks for reading! We'd love to hear your thoughts — questions, feedback, and ideas all welcome.

📄 Paper: https://t.co/S7wLTAs66h
🌐 Project page: https://t.co/ONPhVbk8US

@orlitany @s_elflein @IdoSobol

[Daniel Gilo @danielgilo1]

Conditional diffusion/flow models often produce outputs inconsistent with the very signal conditioning them. The error is easily measurable, yet models are never trained to act on it.

In FlowBender (now on arXiv), we train the model to correct its own errors. 🧵

[Daniel Gilo @danielgilo1]

Finally, is FlowBender just automated gradient guidance? We show the answer is No: 80% of the learned correction is orthogonal to the error gradient. The gradient is utilized — but as a feature driving a non-linear policy, not a scalar weight. No guidance scheme can express this.

[Or Litany @orlitany]

Excited to share our new paper: VideoMDM 📢

We propose a principled framework for training 3D motion diffusion models (e.g. MDM), using only 2D supervision from monocular videos -- no 3D ground truth required.

Project: https://t.co/8Ux0SYRTcu
Paper: https://t.co/8KMgndPWkH
🧵

[Or Litany @orlitany]

This morning @CVPR: [poster #35] we explain how to — consistent edits of multiview Images without training
@danielgilo1 and I will be there to walk you through the method

[Or Litany @orlitany]

This afternoon @CVPR: Realiz3D [poster #526]
Come learn how to make image diffusion models 3D aware without leaking synthetic appearance.

[Ido Sobol @IdoSobol]

Tired of 3D generations that look synthetic? They don’t have to.
Excited to share that Realiz3D is accepted to #CVPR2026 🎉
A framework for training diffusion models that are 3D-consistent, controllable, and photorealistic.
🔗 Project page: https://t.co/8UDsFk6N1A
🎥 Teaser ↓

[Or Litany @orlitany]

Excited to share our new paper, accepted to CVPR 2026: Instruct-Mix2Mix! 📢
We tackle the challenge of sparse multi-view editing: modifying a scene given only a few images (e.g., 4-8) via text instructions.
The code is available now.
Project page: https://t.co/61qC0Q4JJu
🧵