FuseReg: Regularization Bridges Reconstruction-Generation Gap in Representation Autoencoders

USC PSI Lab proposed FuseReg, a layer-fusion regularization that resolves the reconstruction-generation trade-off in representation autoencoders, reducing gFID by 29% while balancing pixel-level and semantic feature preferences.

2026-09-28 ~ 2026-09-29 · 2 related posts

• FuseReg: Layer-Fusion Regularization Cuts gFID up to 29% in Representation Autoencoders — USC-PSI-Lab · 2026-09-28
• FuseReg: Regularizing Layer Fusion to Close the Reconstruction-Generation Gap in RAEs — _akhaliq · 2026-09-29