Generating Images of Rare Concepts Using Pre-trained Diffusion Models

Keyphrases

• Diffusion Model 100%
• Generated Image 100%
• Training Data 66%
• Performance Improvement 33%
• Superior Performance 33%
• Complex Formation 33%
• Tail Distribution 33%
• High Quality Images 33%
• Long Tail 33%
• Failure Mode 33%
• Recognition Benchmark 33%
• Web of Data 33%
• Structured Concept 33%
• Hand Image 33%
• Hand Palm 33%
• Noise Space 33%
• Few-shot Classification 33%
• Semantic Data Augmentation 33%
• Rare Object 33%
• Text-to-image Models 33%
• Text-to-image Diffusion Model 33%
• Insufficient Training Data 33%

Computer Science

• Diffusion Model 100%
• Training Data 75%
• Superior Performance 25%
• Image Quality 25%
• Data Augmentation 25%
• Structured Concept 25%

Engineering

• Diffusion Model 100%
• Limitations 50%
• Image Model 25%
• Common Mode Failure 25%