Research map / Image generation / Fast diffusion sampling
Factored Classifier-Free Guidance
Tian Xia, Fabio De Sousa Ribeiro, Rajat R Rasal, Avinash Kori, Raghav Mehta, Ben Glocker
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace global CFG with factored guidance (separate weights for affected and invariant attributes) | DDIM counterfactual diffusion · CelebA-HQ do(Smiling) | off-target Male AUC change: +3.0 (SA-DCG) -> -1.5 (FCFG), target +12.9 -> +13.1 | Improved |
| use FCFG instead of global CFG | conditional diffusion counterfactuals · MIMIC-CXR do(sex) | off-target AUC change: +10.7 (CFG) -> +0.6 (FCFG) | Improved |
| use FCFG instead of global CFG | conditional diffusion counterfactuals · MIMIC-CXR do(finding) | target AUC change: +18.8 target, off-target +0.6 | Improved |
| increase global CFG weight (ablation) | CFG counterfactual generation · CelebA-HQ, EMBED, MIMIC-CXR | off-target attribute effectiveness (Delta) | Hurt |
| use FCFG instead of global CFG for reversibility | CFG counterfactual generation · CelebA-HQ do(Smiling) | MAE and LPIPS reversibility | Improved |
| use FCFG instead of global CFG | counterfactual generation · CelebA-HQ | FID | Improved |
| integrate FCFG with CFG++ and APG | CFG++, APG · counterfactual generation | attribute amplification | Improved |
| classifier-free dropout probability 0.5 instead of 0.2 (ablation) | diffusion model training · counterfactual generation | identity preservation | Improved |
| DDIM sampling with 1,000 steps instead of shorter schedules (ablation) | counterfactual sampling · counterfactual evaluation | identity preservation | Improved |
Closest papers by meaning
- Counterfactual Generative Modeling with Variational Causal Inference (ICLR 2025 · similarity 0.89)
- Benchmarking Counterfactual Image Generation (NeurIPS 2024 · similarity 0.88)
- Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterations (NeurIPS 2025 · similarity 0.88)
- Counterfactual Image Editing (ICML 2024 · similarity 0.87)
- Towards Characterizing Domain Counterfactuals for Invertible Latent Causal Models (ICLR 2024 · similarity 0.87)
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
