Research map / Image generation / Autoregressive and latent generation
Controllable and Compositional Generation with Latent-Space Energy-Based Models
Weili Nie, Arash Vahdat, Anima Anandkumar
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 ODE sampler in latent space instead of LD sampler (needs fewer steps at similar quality) (ablation) | LACE · CIFAR-10 hyperparameter grid | min average steps for ACC>=95%, FID<=10: 200 -> 41.8 | Improved |
| use ODE sampler instead of LD sampler (best tuned performance) (ablation) | LACE · CIFAR-10 | ACC-FID trade-off | No clear effect |
| use ODE sampler instead of PC sampler (ablation) | LACE · FFHQ glasses | ACC (glasses): 0.943 -> 0.998 | Improved |
| train classifier in w-space rather than z-space or pixel space (ablation) | LACE · CIFAR-10 | ACC/FID | Improved |
| use first-order Euler solver instead of adaptive dopri5 (ablation) | LACE-ODE · CIFAR-10 | ACC/FID | Hurt |
| LACE-ODE latent EBM vs StyleFlow | FFHQ sequential editing · FFHQ yaw_smile_age_glasses | DES: 0.569 -> 0.735 | Improved |
| LACE-ODE vs StyleFlow on unseen attribute combinations | FFHQ zero-shot · FFHQ zero-shot (b) smile | ACC: 0.408 -> 0.906 | Improved |
| LACE-ODE vs StyleFlow on image quality | FFHQ glasses · FFHQ glasses | FID: 43.55 -> 24.52 | Improved |
Closest papers by meaning
- Reduce, Reuse, Recycle: Compositional Generation with Energy-Based Diffusion Models and MCMC (ICML 2023 · similarity 0.88)
- Learning Energy-Based Generative Models via Coarse-to-Fine Expanding and Sampling (ICLR 2021 · similarity 0.88)
- Guiding Energy-based Models via Contrastive Latent Variables (ICLR 2023 · similarity 0.87)
- Compositional Visual Generation with Energy Based Models (NeurIPS 2020 · similarity 0.87)
- Compositional Sculpting of Iterative Generative Processes (NeurIPS 2023 · similarity 0.87)
- Using latent space regression to analyze and leverage compositionality in GANs (ICLR 2021 · 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.
