Research map / Image generation / Autoregressive and latent generation
Distribution Matching Variational AutoEncoder
Sen Ye, Jianning Pei, Mengde Xu, Shuyang Gu, Chunyu Wang, Liwei Wang, Han Hu
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 DINO distribution matching instead of per-sample VAE KL (ablation) | DMVAE · ImageNet 256 reference study | gFID-5k: 27.3 -> 13.1 | Improved |
| use DINO instead of ResNet supervised features as reference (ablation) | DMVAE · ImageNet 256 reference study | gFID-5k: 18.6 -> 13.1 | Improved |
| use DINO instead of SigLIP text features as reference (ablation) | DMVAE · ImageNet 256 reference study | gFID-5k: 26.8 -> 13.1 | Improved |
| use DINO instead of standard Gaussian reference (ablation) | DMVAE · ImageNet 256 reference study | gFID-5k: 26.6 -> 13.1 | Improved |
| use DINO instead of standard Gaussian reference (ablation) | DMVAE · ImageNet 256 reference study | PSNR: 27.4 -> 21.8 | Hurt |
| use full DINO instead of 10-class sub-sampled DINO (ablation) | DMVAE · ImageNet 256 reference study | gFID-5k: 37.9 -> 13.1 | Improved |
| raise lambda_DM from 1 to 10 (ablation) | DMVAE · ImageNet 256 ablation | gFID-5k: 16.7 -> 13.1 | Improved |
| raise lambda_DM from 10 to 100 (ablation) | DMVAE · ImageNet 256 ablation | PSNR: 22.1 -> 19.6 | Hurt |
| use larger XL score network instead of L (ablation) | DMVAE · ImageNet 256 ablation | gFID-5k: 13.1 -> 12.5 | Improved |
| use small score network instead of L (ablation) | DMVAE · ImageNet 256 ablation | gFID-5k: 13.1 -> 13.9 | Hurt |
| apply CFG weight 3.0 in score matching (ablation) | DMVAE · ImageNet 256 ablation | gFID-5k: 13.1 -> 11.6; rFID 0.44 -> 1.02 | Mixed |
| anneal timestep range [0,1] to [0,0.5] (ablation) | DMVAE · ImageNet 256 ablation | gFID-5k: 13.1 -> 12.7 | Improved |
Closest papers by meaning
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.89)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.89)
- CV-VAE: A Compatible Video VAE for Latent Generative Video Models (NeurIPS 2024 · similarity 0.88)
- A Contrastive Learning Approach for Training Variational Autoencoder Priors (NeurIPS 2021 · similarity 0.88)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · similarity 0.88)
- Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation (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.
