AI Paper Atlas

Research map / 3D vision / Gaussian splatting

CloDS: Visual-Only Unsupervised Cloth Dynamics Learning in Unknown Conditions

Yu-Liang Zhan, Jian Li, Wenbing Huang, Yang Liu, Hao Sun

ICLR 2026 · 3D vision › Gaussian splatting · approach: Monocular dynamic scene reconstruction

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
train CloDS on videos with 50 extra video trajectories on top of MGN* mesh pretrainingCloDS (MGN) · FLAGSIMPLE unviewed interpolationrollout RMSE (Avg.): 0.1460 -> 0.1399Improved
train on all videos (CloDS**) versus MGN with full mesh supervisionCloDS · FLAGSIMPLE viewed interpolationrollout RMSE (Avg.): 0.1286 (MGN) vs 0.1294 (CloDS**)No clear effect
use fewer videos (CloDS* 50 videos vs CloDS 50 videos plus meshes) for unviewed trajectoriesCloDS* · FLAGSIMPLE unviewed interpolationrollout RMSE (Avg.): 0.1399 (CloDS) vs 0.1428 (CloDS*)Hurt
replace GaMeS fixed opacity with dual-position opacity modulation (SMGS) (ablation)mesh Gaussian splatting · Dynamic scene novel view synthesisPSNR (dB): 33.0249 -> 36.2368Improved
add world-space and mesh-space coordinates to opacity in mesh extraction (ablation)SMGS · Mesh extraction stage ablationmesh error over timeImproved
render videos through 3D mesh prediction instead of 2D video prediction (CloDS vs SimVP)CloDS · Cloth video predictionPSNR (dB): 25.4770 -> 26.6207Improved
train on multiple trajectories and test on unseen ones (CloDS vs SimVP)CloDS · Cloth video prediction (generalization)PSNR (dB): 14.2391 -> 25.2380Improved
swap MGN for HCMT, DHMP or BSMS-GNN backbone and add video training (ablation)CloDS · Neural simulator ablationrollout RMSEImproved
add noise to initial mesh (Gaussian, translation, scaling) (ablation)CloDS* · FLAGSIMPLE unviewed interpolationrollout RMSE d_avg t<=300: 0.1437 -> 0.1495 (scaling noise)Hurt
add Gaussian noise to initial mesh (ablation)CloDS* · FLAGSIMPLE viewed interpolationrollout RMSE d_avg t<=300: 0.1313 -> 0.1362Hurt

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