Research map / Image generation / Autoregressive and latent generation
The Universal Normal Embedding
Chen Tasker, Roy Betser, Eyal Gofer, Meir Yossef Levi, Guy Gilboa
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 DDIM-inverted diffusion noise instead of encoder embeddings for linear attribute probing | SD 1.5 / SD 2.1 / LCM noise latents · CelebA 40 attributes | attribute classification accuracy | Mixed |
| linearly map generative noise latents into encoder spaces and apply encoder-trained classifiers | SD 1.5 / SD 2.1 / LCM · CelebA attribute prediction | accuracy drop (pp): 0.20 / 0.14 / 0.00 pp drop (CLIP B16) | No clear effect |
| orthogonalize the edit direction against a spurious attribute direction (ablation) | linear latent editing in DDIM-inverted SD 1.5 noise · CelebA edits | qualitative attribute disentanglement | Improved |
| compare generative noise latents with encoder embeddings for Gaussianity of random projections | SD 1.5 vs DINOv3 · Anderson-Darling test, 5,000 projections | AD % Gaussian: 84.48 (DINOv3) -> 96.00 (SD 1.5) | Improved |
| build shared latent space from intersection of multiple encoder and generator latents instead of using a single PCA-reduced space | GCCA shared spaces X1-X5 · CelebA linear probe at 16 dimensions | attribute classification accuracy | No clear effect |
| reduce shared space dimension below 16 (ablation) | shared spaces X1-X5 · CelebA attribute classification | attribute classification accuracy | Hurt |
Closest papers by meaning
- Both Semantics and Reconstruction Matter: Making Representation Encoders Ready for Text-to-Image Generation and Editing (ICML 2026 · similarity 0.89)
- End-to-End Training for Unified Tokenization and Latent Denoising (NeurIPS 2026 · similarity 0.87)
- Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think (NeurIPS 2025 · similarity 0.87)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.87)
- Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective (NeurIPS 2024 · similarity 0.86)
- NoiseCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions in Diffusion Models (CVPR 2024 · similarity 0.86)
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.
