AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Variable-Length Tokenization via Learnable Global Merging for Diffusion Transformers

Dong Hoon Lee, Seunghoon Hong

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers · 0 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
replace nested dropout with learnable global merging for length modulation (ablation)variable-length tokenizer + LightningDiT-B (joint) · ImageNet 256x256, 32 tokensgFID: 5.21 -> 3.19Improved
replace nested dropout with global merging (ablation)variable-length tokenizer · ImageNet 256x256 cross-lengthCKNNA: 0.34 -> 0.76 averageImproved
train one jointly across lengths vs separate models per length (nested dropout) (ablation)nested dropout latents · ImageNet 256x256, 128 tokensgFID: 2.94 -> 4.69Hurt
joint instead of separate training with learnable global merging (ablation)global merging latents · ImageNet 256x256, 128 tokensgFID: 2.48 -> 2.79Hurt
add length-specific LoRA post-training (10 epochs) (ablation)jointly trained DiT on merged latents · ImageNet 256x256, 128 tokensgFID: 2.79 -> 2.52Improved
add LoRA at 32 tokens (ablation)jointly trained DiT on merged latents · ImageNet 256x256, 32 tokensgFID: 3.19 -> 3.04Improved
add global (data-independent) merging with proportional attention and merged positional embeddings over ToMe-style merging (ablation)DiT-B · ImageNet 256x256, 128 tokensgFID: 4.89 -> 4.27Improved
add learnable embeddings to global merging (ablation)DiT-B · ImageNet 256x256, 128 tokensgFID: 4.27 -> 3.21Improved
add alignment loss to learnable global merging (ablation)DiT-B · ImageNet 256x256, 128 tokensgFID: 3.21 -> 2.79Improved
use separate positional embeddings per token count instead of merged ones (ablation)DiT · ImageNet 256x256gFIDNo clear effect
use data-dependent ToMe merging instead of global merging for reconstruction (ablation)tokenizer · ImageNet 256x256 reconstructionrFID/PSNRNo clear effect
more diffusion sampling steps (25 to 100) for our method (ablation)Ours · ImageNet 256x256, 32 tokensgFID: 2.89 -> 2.54Improved

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