Factored Classifier-Free Guidance

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Xia, Tian, Ribeiro, Fabio De Sousa, Rasal, Rajat R, Kori, Avinash, Mehta, Raghav, Glocker, Ben
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913099890556928
author Xia, Tian
Ribeiro, Fabio De Sousa
Rasal, Rajat R
Kori, Avinash
Mehta, Raghav
Glocker, Ben
author_facet Xia, Tian
Ribeiro, Fabio De Sousa
Rasal, Rajat R
Kori, Avinash
Mehta, Raghav
Glocker, Ben
contents Counterfactual generation aims to simulate realistic hypothetical outcomes under causal interventions. Diffusion models have emerged as a powerful tool for this task, combining DDIM inversion with conditional generation and classifier-free guidance (CFG). In this work, we identify a key limitation of CFG for counterfactual generation: it prescribes a global guidance scale for all attributes, leading to significant spurious changes in inferred counterfactuals. To mitigate this, we propose Factored Classifier-Free Guidance (FCFG), a flexible and model-agnostic guidance technique that enables attribute-wise control following a causal graph. FCFG complements recent advances in classifier-free guidance and can be seamlessly extended to advanced guidance schemes such as CFG++ and APG. Our experiments demonstrate that FCFG significantly improves the axiomatic soundness of inferred counterfactuals across both natural and medical image datasets, mitigating spurious amplification effects, and enhancing counterfactual reversibility.
format Preprint
id arxiv_https___arxiv_org_abs_2506_14399
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Factored Classifier-Free Guidance
Xia, Tian
Ribeiro, Fabio De Sousa
Rasal, Rajat R
Kori, Avinash
Mehta, Raghav
Glocker, Ben
Computer Vision and Pattern Recognition
Artificial Intelligence
Counterfactual generation aims to simulate realistic hypothetical outcomes under causal interventions. Diffusion models have emerged as a powerful tool for this task, combining DDIM inversion with conditional generation and classifier-free guidance (CFG). In this work, we identify a key limitation of CFG for counterfactual generation: it prescribes a global guidance scale for all attributes, leading to significant spurious changes in inferred counterfactuals. To mitigate this, we propose Factored Classifier-Free Guidance (FCFG), a flexible and model-agnostic guidance technique that enables attribute-wise control following a causal graph. FCFG complements recent advances in classifier-free guidance and can be seamlessly extended to advanced guidance schemes such as CFG++ and APG. Our experiments demonstrate that FCFG significantly improves the axiomatic soundness of inferred counterfactuals across both natural and medical image datasets, mitigating spurious amplification effects, and enhancing counterfactual reversibility.
title Factored Classifier-Free Guidance
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2506.14399