Contrastive CFG: Improving CFG in Diffusion Models by Contrasting Positive and Negative Concepts
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910716189999104 |
|---|---|
| author | Chang, Jinho Chung, Hyungjin Ye, Jong Chul |
| author_facet | Chang, Jinho Chung, Hyungjin Ye, Jong Chul |
| contents | As Classifier-Free Guidance (CFG) has proven effective in conditional diffusion model sampling for improved condition alignment, many applications use a negated CFG term to filter out unwanted features from samples. However, simply negating CFG guidance creates an inverted probability distribution, often distorting samples away from the marginal distribution. Inspired by recent advances in conditional diffusion models for inverse problems, here we present a novel method to enhance negative CFG guidance using contrastive loss. Specifically, our guidance term aligns or repels the denoising direction based on the given condition through contrastive loss, achieving a nearly identical guiding direction to traditional CFG for positive guidance while overcoming the limitations of existing negative guidance methods. Experimental results demonstrate that our approach effectively removes undesirable concepts while maintaining sample quality across diverse scenarios, from simple class conditions to complex and overlapping text prompts. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2411_17077 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Contrastive CFG: Improving CFG in Diffusion Models by Contrasting Positive and Negative Concepts Chang, Jinho Chung, Hyungjin Ye, Jong Chul Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition As Classifier-Free Guidance (CFG) has proven effective in conditional diffusion model sampling for improved condition alignment, many applications use a negated CFG term to filter out unwanted features from samples. However, simply negating CFG guidance creates an inverted probability distribution, often distorting samples away from the marginal distribution. Inspired by recent advances in conditional diffusion models for inverse problems, here we present a novel method to enhance negative CFG guidance using contrastive loss. Specifically, our guidance term aligns or repels the denoising direction based on the given condition through contrastive loss, achieving a nearly identical guiding direction to traditional CFG for positive guidance while overcoming the limitations of existing negative guidance methods. Experimental results demonstrate that our approach effectively removes undesirable concepts while maintaining sample quality across diverse scenarios, from simple class conditions to complex and overlapping text prompts. |
| title | Contrastive CFG: Improving CFG in Diffusion Models by Contrasting Positive and Negative Concepts |
| topic | Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2411.17077 |