Research map / Image generation / Autoregressive and latent generation
Decoupling Global and Local Representations via Invertible Generative Flows
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Condition every flow coupling layer on z rather than using z only in the prior Gaussian (ablation) | VAE-flow model · CIFAR-10 | BPD: 3.31 -> 3.27 | Helped |
| Condition every flow coupling layer on z rather than using z only in the prior Gaussian (ablation) | VAE-flow model · CIFAR-10 | FID: 43.34 -> 37.52 | Helped |
| Replace exponential scale parameterization with the bounded tanh scale parameterization (ablation) | Affine coupling layers · Image benchmarks | density estimation | Helped |
| Replace original Glow organization with the refined flow architecture (ablation) | Glow · CIFAR-10 | bits/dim: 3.35 -> 3.33 | Helped |
| Add the VAE encoder and condition the refined Glow decoder on global latent z | Glow: refined · CIFAR-10 | BPD: 3.31 -> 3.27 | Helped |
| Add the VAE encoder and condition the refined Glow decoder on global latent z | Glow: refined · CIFAR-10 | FID: 43.34 -> 37.52 | Helped |
Closest papers by meaning
- Variational Lossy Autoencoder (ICLR 2017 · similarity 0.89)
- Deterministic training of generative autoencoders using invertible layers (ICLR 2023 · similarity 0.89)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.88)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.87)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · similarity 0.86)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · 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.