Research map / Image generation / Autoregressive and latent generation
Unsupervised Representation Learning from Pre-trained Diffusion Probabilistic Models
Zijian Zhang, Zhou Zhao, Zhijie Lin
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 |
|---|---|---|---|
| adopt PDAE (pre-trained DPM plus mean-shift estimator) instead of Diff-AE | autoencoding with inferred x_T · CelebA-HQ reconstruction | SSIM: 0.991 -> 0.994 | Improved |
| adopt PDAE instead of Diff-AE | autoencoding with random x_T · CelebA-HQ reconstruction | LPIPS: 0.073 -> 0.098 | Hurt |
| adopt PDAE instead of Diff-AE | autoencoding with random x_T · CelebA-HQ reconstruction | SSIM: 0.677 -> 0.689 | Improved |
| reuse pre-trained DPM encoder and redesign loss weighting (PDAE vs Diff-AE) | representation learning · training throughput and convergence | imgs/sec/A100: 75.41 -> 81.57 | Improved |
| use PDAE instead of Diff-AE for few-shot conditional generation | CelebA64 few-shot · Binary Male | FID: 11.52 -> 11.21 | Improved |
| use PDAE instead of Diff-AE for few-shot conditional generation | CelebA64 few-shot · Binary Blond | FID: 16.10 -> 16.96 | Hurt |
| use PDAE instead of Diff-AE for few-shot conditional generation | CelebA64 PU few-shot · PU Blond | FID: 7.01 -> 6.34 | Improved |
| use simple loss weighting lambda_t=1 instead of SNR-based redesigned weighting (ablation) | PDAE training · training stability | convergence | Improved |
| apply PDAE gradient estimator only in the first 70% of sampling steps for unconditional sampling (ablation) | pre-trained DPM sampling · unconditional FFHQ/Horse/Bedroom/CelebA64 | FID | Improved |
Closest papers by meaning
- LMD: Faster Image Reconstruction with Latent Masking Diffusion (AAAI 2024 · similarity 0.90)
- Deconstructing Denoising Diffusion Models for Self-Supervised Learning (ICLR 2025 · similarity 0.89)
- Diffusion Bridge AutoEncoders for Unsupervised Representation Learning (ICLR 2025 · similarity 0.89)
- Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models (ICLR 2025 · similarity 0.88)
- Slight Corruption in Pre-training Data Makes Better Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Learning Diffusion Models with Flexible Representation Guidance (NeurIPS 2025 · similarity 0.88)
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.
