JEDI: The Force of Jensen-Shannon Divergence in Disentangling Diffusion Models
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866915405405093888 |
|---|---|
| author | Bill, Eric Tillmann Simsar, Enis Hofmann, Thomas |
| author_facet | Bill, Eric Tillmann Simsar, Enis Hofmann, Thomas |
| contents | We introduce JEDI, a test-time adaptation method that enhances subject separation and compositional alignment in diffusion models without requiring retraining or external supervision. JEDI operates by minimizing semantic entanglement in attention maps using a novel Jensen-Shannon divergence based objective. To improve efficiency, we leverage adversarial optimization, reducing the number of updating steps required. JEDI is model-agnostic and applicable to architectures such as Stable Diffusion 1.5 and 3.5, consistently improving prompt alignment and disentanglement in complex scenes. Additionally, JEDI provides a lightweight, CLIP-free disentanglement score derived from internal attention distributions, offering a principled benchmark for compositional alignment under test-time conditions. Code and results are available at https://ericbill21.github.io/JEDI/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_19166 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | JEDI: The Force of Jensen-Shannon Divergence in Disentangling Diffusion Models Bill, Eric Tillmann Simsar, Enis Hofmann, Thomas Computer Vision and Pattern Recognition Machine Learning We introduce JEDI, a test-time adaptation method that enhances subject separation and compositional alignment in diffusion models without requiring retraining or external supervision. JEDI operates by minimizing semantic entanglement in attention maps using a novel Jensen-Shannon divergence based objective. To improve efficiency, we leverage adversarial optimization, reducing the number of updating steps required. JEDI is model-agnostic and applicable to architectures such as Stable Diffusion 1.5 and 3.5, consistently improving prompt alignment and disentanglement in complex scenes. Additionally, JEDI provides a lightweight, CLIP-free disentanglement score derived from internal attention distributions, offering a principled benchmark for compositional alignment under test-time conditions. Code and results are available at https://ericbill21.github.io/JEDI/. |
| title | JEDI: The Force of Jensen-Shannon Divergence in Disentangling Diffusion Models |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2505.19166 |