AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Aligning Visual Foundation Encoders to Tokenizers for Diffusion Models

Bowei Chen, Sai Bi, Hao Tan, He Zhang, Tianyuan Zhang, Zhengqi Li, Yuanjun Xiong, Jianming Zhang, Kai Zhang

ICLR 2026 · Image generation › Autoregressive and latent generation · approach: Hierarchical VQ-VAE generation

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
Align pretrained DINOv2 encoder instead of semantic regularization (VA-VAE)Tokenizer · ImageNet 256 f16d32, w/ CFGgFID: 3.13 -> 2.17Improved
Align pretrained encoder instead of semantic regularizationTokenizer · ImageNet 256 unconditionalgFID: 19.12 -> 11.80Improved
Add semantic preservation loss in stage 2 (ablation)AlignTok · ImageNet 256gFID: 3.05 -> 2.19Improved
Raise semantic preservation weight to 5 or 10 (ablation)AlignTok · ImageNet 256rFID / gFID: rFID 0.36 -> 0.59; gFID 2.19 -> 2.59Mixed
Apply semantic loss before the adapter instead of on latent (ablation)AlignTok · ImageNet 256gFID: 2.19 -> 2.83Hurt
Replace L2 semantic loss with cosine loss (ablation)AlignTok · ImageNet 256gFID: 2.19 -> 2.23No clear effect
Use LoRA instead of full fine-tuning in stage 2 (ablation)AlignTok · ImageNet 256rFID / gFID: rFID 0.36 -> 1.35; gFID 2.19 -> 2.97Hurt
Add EMA in stage 2 (ablation)AlignTok · ImageNet 256gFID: 2.24 -> 2.19Improved
Use a larger decoder (ablation)AlignTok · ImageNet 256gFID: 2.19 -> 2.52Hurt
Add stage 2 perceptual alignment to stage 1 (ablation)AlignTok · ImageNet 256rFID: 1.63 -> 0.36Improved
Add stage 3 decoder refinement (ablation)AlignTok · ImageNet 256rFID / gFID: rFID 0.36 -> 0.26; gFID 2.19 -> 2.17Improved
Use DINOv2 instead of MAE or SigLIP 2 (ablation)AlignTok · ImageNet 256gFID: MAE 3.12, SigLIP 2 2.22 -> 2.19Improved

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