AI Paper Atlas

Research map / Video / Video generation

LeanVAE: An Ultra-Efficient Reconstruction VAE for Video Diffusion Models

Yu Cheng, Fajie Yuan

ICCV 2025 · Video › Video generation · approach: Efficient video diffusion models · 13 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
use ISTA-Net+ compressed sensing instead of a two-linear-layer autoencoder for the latent channel bottleneck (ablation)LeanVAE · DAVIS (4 latent channels, 250k steps)PSNR / LPIPS: 0.39 dB in PSNR and 0.018 in LPIPSImproved
remove LayerNorm on pixel patches (no patch normalization) (ablation)LeanVAE · DAVIS (4 latent channels, 250k steps)PSNR / LPIPS: 3.27 dB in PSNR and 0.013 in LPIPSImproved
process low and high frequency separately then fuse (Variant 2) versus joint processing (ablation)LeanVAE · DAVIS (4 latent channels, 250k steps)PSNR / SSIMImproved
add Haar wavelet transform on input (Variant 2 vs Variant 3 without wavelet) (ablation)LeanVAE · DAVIS (4 latent channels, 250k steps)reconstruction qualityImproved
increase latent channels from 4 to 16 in generation (ablation)Ours-Latte · UCF101 generationFVD: 164.45 -> 175.33Hurt
increase latent channels from 4 to 16 in generation (ablation)Ours-Latte · SkyTimelapse generationFVD: 49.59 -> 95.15Hurt
replace SD VAE with LeanVAE (4 chn) in LatteLatte-XL · UCF101 generationFVD: 477.97 -> 164.45Improved
replace SD VAE with LeanVAE (4 chn) in LatteLatte-XL · SkyTimelapse generationFVD: 59.82 -> 49.59Improved
replace SD VAE with LeanVAE in Latte for training throughputLatte-XL · training speedsamples/sec: 1.60 -> 6.64Improved
use temporal tiling with chunksize 5 on VidTok versus full inferenceVidTok · DAVIS 17x256x256rFVD: 358.28 -> 342.44Mixed
use temporal tiling with chunksize 5 on LeanVAE versus full inferenceLeanVAE · DAVIS 17x256x256rFVD: 322.56 vs 322.46No clear effect
use LeanVAE instead of VidTok at 768x768 resolutionLeanVAE · 768x768 17-frame videoinference time: 20.26 seconds -> 0.46 secondsImproved

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