OmnimatteZero: Fast Training-free Omnimatte with Pre-trained Video Diffusion Models

Keyphrases

• Background Reconstruction 100%
• Individual Objects 100%
• Training-free 100%
• Fast Training 100%
• Video Diffusion Model 100%
• Superior Performance 50%
• Spatial Attention 50%
• Diffusion Process 50%
• Extensive Training 50%
• Real-time Performance 50%
• Inpainting 50%
• Image Inpainting 50%
• Temporal Attention 50%
• Zero-shot 50%
• Object Removal 50%
• Background Objects 50%
• Temporally Consistent 50%
• Self-supervised Optimization 50%
• Self-attention Map 50%
• Inpainting Technique 50%
• Guidance Module 50%
• Video Object Removal 50%
• Attention Guidance 50%

Computer Science

• Superior Performance 100%
• Real Time Performance 100%
• Diffusion Process 100%
• Image Inpainting 100%
• Extensive Training 100%
• Zero-Shot Learning 100%
• Self-Attention 100%