DogFit: Domain-guided Fine-tuning for Efficient Transfer Learning of Diffusion Models

Keyphrases

• Transfer Learning 100%
• Diffusion Model 100%
• Parameter Transfer Learning 100%
• Guidance Method 66%
• Domain-aware 66%
• Target Domain 33%
• Optimal Placement 33%
• Tuning Process 33%
• Source Model 33%
• Guidance Mechanism 33%
• High Computational Cost 33%
• New Domains 33%
• Quality Stability 33%
• Image Fidelity 33%
• Fine-tuning Strategy 33%
• Training Loss 33%
• Scheduling Strategy 33%
• Poor Generalization 33%
• Forward Pass 33%
• Optimal Timing 33%
• Computational Overhead 33%
• Strength Value 33%
• Training Stability 33%
• Diffusion Transfer 33%
• Conditioning Mechanism 33%
• Late Start 33%
• Sample Diversity 33%
• Marginal Estimates 33%

Computer Science

• Transfer Learning 100%
• Computational Cost 25%
• Scheduling Strategy 25%
• Vertebral Column 25%