AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

LayoutFlow: Flow Matching for Layout Generation

Julian Jorge Andrade Guerreiro, Naoto Inoue, Kento Masui, Mayu Otani, Hideki Nakayama

ECCV 2024 · Image generation › Fast diffusion sampling · approach: Flow matching distillation · 19 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
replace diffusion (DDPM sampling) with flow matching using same architecture (ablation)LayoutFlow · RICO Un-GenFID: 34.89 -> 2.37Improved
replace diffusion (DDIM sampling) with flow matching using same architecture (ablation)LayoutFlow · RICO Un-GenFID: 3.18 -> 2.37Improved
add L1 geometric regularization to flow matching loss (lambda 0.2) (ablation)LayoutFlow · RICO Un-GenAlignment: 0.194 -> 0.150Improved
increase L1 regularization weight from 0.2 to 0.4 (ablation)LayoutFlow · RICO Un-GenFID: 2.37 -> 2.60Hurt
use L1 loss only without flow matching loss (ablation)LayoutFlow · RICO Un-GenFID: 2.37 -> 26.64Hurt
use uniform instead of Gaussian initial distribution (ablation)LayoutFlow · RICO Un-GenFID: FID 2.37 -> 2.61; Alignment 0.150 -> 0.115Mixed
use sine/cosine or sine training trajectory instead of linear (ablation)LayoutFlow · RICO Un-GenFID: 2.37 -> 2.33 (sine/cosine), 2.48 (sine)No clear effect
train with masking strategy instead of imposing conditions on trajectory at inference (ablation)LayoutFlow · RICO Gen-TypeSizeFID: 174.66 -> 1.03Improved
train with masking strategy instead of inference-time trajectory conditioning (ablation)LayoutFlow · RICO CompletionFID: 7.96 -> 1.51Improved
use flow matching in place of discrete diffusion in LayoutDiffusion on PubLayNet Un-Genlayout generation · PubLayNet Un-GenFID: 8.63 -> 8.87No clear effect
use flow matching with 50 generation steps and 20 tokens per layoutlayout generation · Inference speedinference time (ms): 3.50ms -> 1.75ms (vs DLT)Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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