AI Paper Atlas

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

Reconciling Visual Perception and Generation in Diffusion Models

Liulei Li, Yi Yang, Wenguan Wang

ICLR 2026 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance

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 generative visual perception learning (ablation)GenRep (LDM) · ObjectNet / ADE20Ktop-1 / mIoU: 45.4 -> 47.8 top-1; 27.8 -> 30.9 mIoUImproved
add semantic-driven generation learning (ablation)GenRep (LDM) · ImageNet 256FID: 13.27 -> 7.45Improved
add gradient alignment (ablation)GenRep (LDM) · ObjectNet / ADE20K / ImageNettop-1 / mIoU / FID: 49.4 -> 51.1 top-1; 31.5 -> 32.5 mIoU; 7.23 -> 6.92 FIDImproved
add generative distillation loss L_gen_distil (ablation)GenRep (LDM) · ObjectNetECE: 0.237 -> 0.208Improved
thin samples with interval k=2 instead of k=1 (ablation)GenRep (LDM) · ObjectNettop-1: 50.3 -> 51.1Improved
increase thinning interval to k=5 (ablation)GenRep (LDM) · ObjectNettop-1: 51.1 -> 48.9Hurt
increase burn-in to m=125 from m=50 (ablation)GenRep (LDM) · ObjectNettop-1: 51.1 -> 47.6Hurt
add importance re-weighting (linear scaling) of samples (ablation)GenRep (LDM) · ObjectNettop-1: 51.1 -> 49.2Hurt
add both noise adjustment and representation alignment (ablation)GenRep (LDM) · ImageNet 256FID: 7.16 -> 6.92Improved
symmetric gradient alignment instead of generation-only alignment (ablation)GenRep (LDM) · ObjectNet / ADE20K / ImageNettop-1 / mIoU / FID: 48.7 -> 50.1 top-1; 30.3 -> 32.5 mIoU; 6.79 -> 6.92 FIDMixed
use larger thinning interval k=4 with fixed 75 samples (ablation)GenRep (LDM) · ObjectNettop-1: 51.1 -> 51.9Improved
task weights wp=0.8, wg=0.2 instead of 0.7/0.3 (ablation)GenRep (LDM) · ObjectNet / ImageNet 256top-1 / FID: 51.1 -> 51.5 top-1; 6.92 -> 7.12 FIDMixed

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