Research map / Image generation / Autoregressive and latent generation
From smooth to sharp: Frequency-Decoupled Latent Optimization for Realistic Image Generation
Tejaswini Medi, Hsien-Yi Wang, Arianna Rampini, Margret Keuper
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 |
|---|---|---|---|
| decouple low and high frequency bands into separate encoders and decoders (ablation) | VA-VAE with wavelet input · ImageNet-1K val reconstruction | reconstruction loss: 0.0125 -> 0.0044 | Improved |
| decouple low and high frequency bands (ablation) | VA-VAE with wavelet input · ImageNet-1K val reconstruction | high-frequency loss: 0.0098 -> 0.0020 | Improved |
| decouple low and high frequency bands (ablation) | VA-VAE with wavelet input · ImageNet-1K val reconstruction | PSNR: 26.01 -> 30.03 | Improved |
| decouple low and high frequency bands (ablation) | VA-VAE with wavelet input · ImageNet-1K val reconstruction | rFID: 0.6509 -> 0.4156 | Improved |
| decouple bands in KL-VAE and MS-VQ-VAE style tokenizers (coupled baselines) (ablation) | KL-VAE with wavelet input · ImageNet-1K val reconstruction | reconstruction loss: 0.1146 -> 0.0044 | Improved |
| use DeBaT tokenizer instead of VA-VAE in LightningDiT | LightningDiT · ImageNet-1K 256 w/o CFG | gFID: 5.14 -> 3.24 | Improved |
| use DeBaT tokenizer instead of VA-VAE in LightningDiT | LightningDiT · ImageNet-1K 256 w/ CFG | gFID: 2.11 -> 1.52 | Improved |
| use DeBaT tokenizer instead of VA-VAE in LightningDiT | LightningDiT · ImageNet-1K 256 w/ CFG | IS: 252.3 -> 317.4 | Improved |
| use DeBaT instead of VA-VAE | LightningDiT · FFHQ 1024 | gFID: 6.92 -> 6.86 | No clear effect |
| use DeBaT instead of VA-VAE | LightningDiT · FFHQ 1024 | PSNR: 33.84 -> 34.27 | Improved |
| use DeBaT instead of LiteVAE wavelet tokenizer | LiteVAE · ImageNet-1K val reconstruction | high-frequency loss: 0.0050 -> 0.0020 | Improved |
| use dual encoder-decoder (DeBaT) instead of VA-VAE | VA-VAE · tokenizer efficiency | parameters (M): 72.60 -> 137.91 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LiteVAE — LiteVAE: Lightweight and Efficient Variational Autoencoders for Latent Diffusion Models (NeurIPS 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Latent Harmony: Synergistic Unified UHD Image Restoration via Latent Space Regularization and Controllable Refinement (NeurIPS 2025 · similarity 0.87)
- Improving the Diffusability of Autoencoders (ICML 2025 · similarity 0.87)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.86)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.86)
- When Worse is Better: Navigating the Compression Generation Trade-off In Visual Tokenization (NeurIPS 2025 · similarity 0.86)
- Epsilon-VAE: Denoising as Visual Decoding (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.
