❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers

📝 The paper and the code are now available here:
https://dvirsamuel.github.io/omnimattezero.github.io/
https://github.com/dvirsamuel/OmnimatteZero

Our Patreon if you wish to support us: https://www.patreon.com/TwoMinutePapers

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Adam Bridges, Benji Rabhan, B Shang, Cameron Navor, Christian Ahlin, Eric T, Fred R, Gordon Child, Juan Benet, Michael Tedder, Owen Skarpness, Richard Sundvall, Ryan Stankye, Steef, Taras Bobrovytsky, Tazaur Sagenclaw, Tybie Fitzhugh, Ueli Gallizzi
 
My research: https://cg.tuwien.ac.at/~zsolnai/

#nvidia

Two Minute Papers's resource offers insights, tutorials, and resources for researchers and enthusiasts interested in computer science and artificial intelligence. Readers can learn about  research papers, breakthroughs, and trends in the field of AI. With concise summaries, analysis, and visualizations, Two Minute Papers provides  guidance and expertise for understanding complex research topics in a digestible format.

Two Minute Papers

NVIDIA's OmnimatteZero enables real-time video object removal at 25 fps without additional AI training. Unlike previous methods that struggle with secondary effects like shadows and reflections, this technique uses pre-trained diffusion models to intelligently copy background information from adjacent frames rather than generating new content. The approach leverages mean temporal attention to identify and remove objects along with their associated effects (shadows, reflections, grass movement) by recognizing elements that move together across frames. While results show slight blurring due to temporal averaging, the method achieves superior performance compared to 2023 and 2025 techniques, with source code promised for early February release.

NVIDIA’s New AI Just Leveled Up Video Editing