AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Rethinking Garment Conditioning in Diffusion-based Virtual Try-On: Decouple, Don't Denoise

Kihyun Na, Jinyoung Choi, Injung Kim

ECCV 2026 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance · 1 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
switch CatVTON from attention-only to full fine-tuning (ablation)CatVTON · VITON-HD 512x384Paired FID: 5.425 -> 5.250No clear effect
add Garment-Free Guidance (garment-free unconditional branch) (ablation)CatVTON full FT · VITON-HD 512x384Paired FID: 5.250 -> 4.538Improved
add Garment-Free Guidance (ablation)CatVTON full FT · VITON-HD 512x384Paired KID: 0.402 -> 0.074Improved
add Decoupled Loss (person-region loss only) (ablation)CatVTON full FT + GFG · VITON-HD 512x384Paired FID: 4.538 -> 4.517Improved
add Clean Latent Anchoring at inference (ablation)CatVTON full FT + GFG + DL · VITON-HD 512x384Paired FID: 4.517 -> 4.438Improved
add Clean Latent Anchoring at inference (ablation)CatVTON full FT + GFG + DL · VITON-HD 512x384Paired KID: 0.068 -> 0.010Improved
add Decoupled Loss to GFG+CLA (ablation)DeCo-VTON · VITON-HD 512x384Unpaired FID: 8.308 -> 8.266Improved
add Decoupled Loss to GFG+CLA (ablation)DeCo-VTON · VITON-HD 512x384Paired KID: 0.004 -> 0.010Hurt
add Decoupled Loss to GFG+CLA (ablation)DeCo-VTON · VITON-HD 1024x768Unpaired FID: 8.988 -> 8.675Improved
add Decoupled Loss to GFG+CLA (ablation)DeCo-VTON · VITON-HD 1024x768Paired FID: 4.919 -> 4.741Improved
use garment-free guidance and anchoring so guidance scale can be raised (ablation)DeCo-VTON vs CatVTON · VITON-HD guidance scale sweepFIDImproved
full fine-tuning instead of attention-only in Voost DiT (reported by Voost)Voost · VITON-HD 1024x768FID: 5.269 -> 6.351Hurt

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