AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

ViTok-v2: Scaling Native Resolution Autoencoders to 5 Billion Parameters

Philippe Hansen-Estruch, Jiahui Chen, Vivek Ramanujan, Orr Zohar, Markos Georgopoulos, Animesh Sinha, Ji Hou, Edgar Schönfeld, Felix Juefei-Xu, Sriram Vishwanath, Ali Thabet

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Dataset condensation and flow estimation

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 LPIPS loss to pixel-only training (ablation)ViTok-v2 5B · ImageNet-1k 256p loss ablation (5B, r=12)rFID: 5.13 -> 0.72Improved
add DINOv3 perceptual tile loss (lambda=1000) to pixel-only training (ablation)ViTok-v2 5B · ImageNet-1k 256p loss ablation (5B, r=12)rFID: 5.13 -> 0.30Improved
add DINOv3 loss (lambda=1000) to pixel-only training (ablation)ViTok-v2 5B · ImageNet-1k 256p loss ablation (5B, r=12)SSIM: .929 -> .914Hurt
add SSIM loss to pixel-only training (ablation)ViTok-v2 5B · ImageNet-1k 256p loss ablation (5B, r=12)rFID: 5.13 -> 4.87Improved
use DINOv3 loss instead of LPIPS (ablation)ViTok-v2 5B · ImageNet-1k 256p loss ablation (5B, r=12)rFID: 0.72 -> 0.30Improved
add LPIPS on top of DINOv3 loss (ablation)ViTok-v2 5B · ImageNet-1k 256p loss ablation (5B, r=12)rFID: 0.30 -> 0.38No clear effect
replace 4-block encoder with a single block (ablation)ViTok-v2 f16 · SSIMSSIM: .768 -> .753No clear effect
replace encoder with a linear projection (ablation)ViTok-v2 f16 · SSIMSSIM: .768 -> .462Hurt
scale f32 model width from Bd4-B to Td4-T (ablation)ViTok-v2 f32 · SSIMSSIM: .432 -> .872Improved
scale decoder from B to T at r=48 (ablation)ViTok-v2 · ImageNet-1k 256prFID: B-to-T rFID gap 2.3 at r=12 grows to 12.2 at r=48Improved
use LayerNorm instead of KL or tanh+noise latent regularization (ablation)ViTok-v2 AE + 450M flow · ImageNet-1k 256pgFID: 4.71 -> 4.73No clear effect
use 5B AE instead of 350M AE for generation (ablation)1.2B flow transformer · ImageNet-22kgFID: 1-2 gFID points betterImproved

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