AI Paper Atlas

Research map / Image generation / Image restoration

High-Fidelity Pluralistic Image Completion With Transformers

Ziyu Wan, Jingbo Zhang, Dongdong Chen, Jing Liao

ICCV 2021 · Image generation › Image restoration · approach: Priors for restoration networks · 316 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
use transformer appearance prior plus guided CNN upsampling instead of PIC's fully convolutional variational modelimage completion · ImageNet, mask 40%-60%FID: 101.278 -> 59.486Improved
use transformer appearance prior plus guided CNN upsampling instead of PIC's fully convolutional variational modelimage completion · ImageNet, mask 20%-40%FID: 47.750 -> 28.818Improved
use transformer appearance prior plus guided CNN upsampling instead of PIC's fully convolutional variational modelimage completion · ImageNet, random masksFID: 59.428 -> 35.842Improved
use proposed method instead of best baseline ECimage completion · FFHQ, mask 20%-40%FID: 14.276 -> 10.995Improved
switch from top-50 sampling to deterministic top-1 sampling (ablation)Ours · FFHQ, mask 20%-40%FID: 10.995 -> 10.515Improved
switch from top-50 sampling to deterministic top-1 sampling (ablation)Ours · FFHQ, mask 40%-60%FID: 20.024 -> 20.843Hurt
switch from top-50 sampling to deterministic top-1 sampling (ablation)Ours · Places2, random masksFID: 25.420 -> 25.985Hurt
switch from top-50 sampling to deterministic top-1 sampling (ablation)Ours · Places2, mask 40%-60%PSNR: 22.215 -> 22.635Improved
remove instance normalization from the upsampling network (ablation)guided upsampling CNN · image completionvisual qualityImproved
add transformer prior versus fully convolutional methods on geometric primitive completionimage completion · pentagram shape completion (ImageNet-trained)qualitative reconstructionImproved
use proposed method versus other baselines in a user studyimage completion · user study ranking, 30 images, 28 participantstop-1 rank ratio: 73.70%Improved

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