AI Paper Atlas

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

Improving Text-guided Object Inpainting with Semantic Pre-inpainting

Yifu Chen, Jingwen Chen, Yingwei Pan, Yehao Li, Ting Yao, Zhineng Chen, Tao Mei

ECCV 2024 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance · 12 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
add reference adapter layer fed with masked-image CLIP features (ablation)Finetuned Stable Inpainting · OpenImages-V6 (segmentation mask)FID / Local FID: FID 5.15 -> 4.24; Local FID 8.80 -> 7.33Improved
add semantic inpainter outputs on top of reference adapter (ablation)CAT-Diffusion · OpenImages-V6 (segmentation mask)FID / Local FID: FID 4.24 -> 3.99; Local FID 7.33 -> 6.32Improved
add semantic inpainter and adapter on top of fine-tuned inpainting model (ablation)Stable Inpainting · OpenImages-V6 (segmentation mask)CLIP score: 0.262 -> 0.263No clear effect
semantic inpainter refines CLIP features of masked regionCAT-Diffusion · 10K imagescosine similarity to ground truth: 0.47 -> 0.65Improved
CAT-Diffusion versus SmartBrushtext-guided object inpainting · OpenImages-V6 (segmentation mask)Local FID: 7.82 -> 6.32Improved
CAT-Diffusion versus SmartBrush on CLIP scoretext-guided object inpainting · OpenImages-V6 (bounding box mask)CLIP score: 0.266 -> 0.265No clear effect
CAT-Diffusion versus Stable Inpainting and SmartBrush zero-shot on MSCOCOtext-guided object inpainting · MSCOCO (bounding box mask)Local FID: relative improvements of 42.1% and 20.7%Improved
CAT-Diffusion versus SD-Inpaint in user studytext-guided object inpainting · OpenImages-V6 user studyVisual coherence (0-5): 2.6 -> 3.6Improved
add semantic inpainter and adapterStable Inpainting · inference time per imageseconds: 1.60 -> 1.84Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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