Research map / Image generation / Fast diffusion sampling
DisDiff: Unsupervised Disentanglement of Diffusion Probabilistic Models
Tao Yang, Yuwang Wang, Yan Lu, Nanning Zheng
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 |
|---|---|---|---|
| use latent diffusion (VQ/KL) instead of image-space DDIM (ablation) | DisDiff · Shapes3D | DCI / FactorVAE score | Improved |
| remove invariant loss L_in from disentangling loss (ablation) | DisDiff-VQ · Shapes3D | disentanglement | Hurt |
| remove variant loss L_va from disentangling loss (ablation) | DisDiff-VQ · Shapes3D | disentanglement | Hurt |
| optimize disentangling loss on both decoder and encoder (ablation) | DisDiff · Shapes3D | disentanglement | Hurt |
| use constant or loss weighting instead of encoder-difference weighting (ablation) | DisDiff · Shapes3D | disentanglement | Hurt |
| use cross-attention instead of AdaGN for condition injection (ablation) | DisDiff · Shapes3D | disentanglement | Hurt |
| remove orthogonal positional embedding or use learnable embedding (ablation) | DisDiff · Shapes3D | disentanglement | Hurt |
| latent number greater than number of ground-truth factors (ablation) | DisDiff · Shapes3D | disentanglement | No clear effect |
| reduce latent number to 3 (fewer than 6 factors) | DisDiff · Shapes3D | DCI: 0.723 -> 0.3279 | Hurt |
| DisDiff vs DisCo on MPI3D | DisDiff-VQ · MPI3D | FactorVAE score: 0.371 -> 0.617 | Improved |
| DisDiff vs DisCo on Cars3D DCI | DisDiff-VQ · Cars3D | DCI: 0.271 -> 0.232 | Hurt |
| DisDiff vs InfoDiffusion on CelebA | DisDiff · CelebA | TAD: 0.299 -> 0.3054 | Improved |
Later papers that compared with this method
3 later accepted papers report DisDiff as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Diffusion Bridge AutoEncoders for Unsupervised Representation Learning (ICLR 2025 · better on 1 of 1 shared result)
- Diffusion Model with Cross Attention as an Inductive Bias for Disentanglement (NeurIPS 2024 · better on 4 of 5 shared results)
- Closed-Loop Unsupervised Representation Disentanglement with $\\beta$-VAE Distillation and Diffusion Probabilistic Feedback (ECCV 2024 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- InfoDiffusion — InfoDiffusion: Representation Learning Using Information Maximizing Diffusion Models (ICML 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Can Diffusion Models Disentangle? A Theoretical Perspective (NeurIPS 2025 · similarity 0.89)
- Diffusion Model with Cross Attention as an Inductive Bias for Disentanglement (NeurIPS 2024 · similarity 0.89)
- Isometric Representation Learning for Disentangled Latent Space of Diffusion Models (ICML 2024 · similarity 0.88)
- DisCo-Diff: Enhancing Continuous Diffusion Models with Discrete Latents (ICML 2024 · similarity 0.88)
- Why DiT Models Underperform as Representation Learners without Long Skip Connections (NeurIPS 2026 · similarity 0.87)
- Learning Diffusion Models with Flexible Representation Guidance (NeurIPS 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.
