AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

DA-VAE: Plug-in Latent Compression for Diffusion via Detail Alignment

Xin Cai, Zhiyuan You, Zhoutong Zhang, Tianfan Xue

CVPR 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
add detail alignment loss to detail channels (ablation)DA-VAE · ImageNet 512 ablationFID-10k: 16.37 -> 9.27Improved
zero-initialize new patch embedder and output layer (ablation)DA-VAE fine-tuning · ImageNet 512 ablationFID-10k: 29.73 -> 9.27Improved
add cosine-annealed loss weight schedule on detail channels (ablation)DA-VAE fine-tuning · ImageNet 512 ablationFID-10k: 9.80 -> 9.27Improved
increase alignment weight from 0.5 to 1.0 (ablation)DA-VAE · ImageNet 512 ablationrFID: 0.47 -> 0.63Hurt
increase alignment weight from 0.5 to 1.0 (ablation)DA-VAE · ImageNet 512 ablationFID-10k: 9.27 -> 9.23No clear effect
increase alignment weight from 0 to 0.1 (ablation)DA-VAE · ImageNet 512 ablationFID-10k: 16.37 -> 9.58Improved
use DA-VAE latent (f32c128) instead of VA-VAE (f16c32)DiT-XL from scratch · ImageNet 512FID-10k: 44.65 -> 31.51Improved
fine-tune with DA-VAE instead of DC-Gen fine-tuningLightningDiT-XL · ImageNet 512 (80 epochs)FID-50k w/ CFG: 2.22 -> 1.68Improved
fine-tune SD3.5M with DA-VAE at 32x32 tokensSD3.5-medium · MJHQ-30K 1024x1024throughput: 0.25 -> 1.03 img/sImproved
fine-tune SD3.5M with DA-VAE at 32x32 tokensSD3.5-medium · MJHQ-30K 1024x1024FID: 10.31 -> 10.91Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy