classifier-free guidance
A technique used in generative models, particularly diffusion models, which allows for improved samples without needing a separate classifier for conditioning. It balances the trade-off between diversity and fidelity in generated outputs by controlling the influence of guidance during the sampling process.
- Adaptive Classifier-Free Guidance via Dynamic Low-Confidence Masking
- Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models
- DICEPTION: A Generalist Diffusion Model for Visual Perceptual Tasks
- DISCO: DISCrete nOise for Conditional Control in Text-to-Image Diffusion Models
- Entropy Rectifying Guidance for Diffusion and Flow Models
- Feedback Guidance of Diffusion Models
- Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling
- HAODiff: Human-Aware One-Step Diffusion via Dual-Prompt Guidance
- MiniMax-Remover: Taming Bad Noise Helps Video Object Removal
- Model-Guided Dual-Role Alignment for High-Fidelity Open-Domain Video-to-Audio Generation
- Normalized Attention Guidance: Universal Negative Guidance for Diffusion Models
- Policy Optimized Text-to-Image Pipeline Design
- Rectified CFG++ for Flow Based Models
- ShortListing Model: A Streamlined Simplex Diffusion for Discrete Variable Generation
- Token Perturbation Guidance for Diffusion Models
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration
- Towards Understanding the Mechanisms of Classifier-Free Guidance
- Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterations
- URLs Help, Topics Guide: Understanding Metadata Utility in LLM Training
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation
- Walking the Schrödinger Bridge: A Direct Trajectory for Text-to-3D Generation