AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Flow to the Mode: Mode-Seeking Diffusion Autoencoders for State-of-the-Art Image Tokenization

Kyle Sargent, Kyle Hsu, Justin Johnson, Li Fei-Fei, Jiajun Wu

ICCV 2025 · Image generation › Autoregressive and latent generation · approach: Hierarchical VQ-VAE generation · 27 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 mode-seeking post-training (Stage 1B) (ablation)FlowMo-Lo · ImageNet-1K tokenizationrFID: 1.10 -> 0.95Improved
add mode-seeking post-training (Stage 1B) (ablation)FlowMo-Hi · ImageNet-1K tokenizationrFID: 0.73 -> 0.56Improved
add mode-seeking post-training (Stage 1B) (ablation)FlowMo-Lo · ImageNet-1K tokenizationPSNR: 21.38 -> 22.07Improved
add mode-seeking post-training (Stage 1B) (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 2.02 -> 1.21Improved
add mode-seeking post-training (Stage 1B) (ablation)FlowMo-Continuous · ImageNet-1K tokenizationrFID: 0.74 -> 0.65Improved
add perceptual loss on 1-step prediction (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 13.86 -> 2.87Improved
train encoder end-to-end rather than with MSE (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 3.82 -> 2.87Improved
use thick-tailed logit-normal noise schedule (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationPSNR: 16.45 -> 20.71Improved
use halved patch size (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 6.39 -> 2.87Improved
use LFQ instead of FSQ (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 3.14 -> 2.87Mixed
use shifted sampler (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 3.42 -> 2.87Improved
add guidance in a limited interval (ablation)FlowMo (fewer params) · ImageNet-1K tokenizationrFID: 3.28 -> 2.87Improved

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