AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Reduce Information Loss in Transformers for Pluralistic Image Inpainting

CVPR 2022 · Architectures and efficiency › Attention and transformers · approach: Transformer block and depth design · 104 citations (OpenAlex)

Check your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's title only (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 PUT50 rather than ICT50pluralistic image inpainting · ImageNet, mask ratio 40-60%FID: 55.598 -> 45.153Helped
use PUTall rather than ICTallpluralistic image inpainting · Places2, mask ratio 10-60%FID: 23.331 -> 24.605Hurt
replace the non-overlapping patch encoder with a convolutional encoder (ablation)PUT · FFHQ, mask ratio 20-40%FID: 12.784 -> 163.610Hurt
use one codebook instead of separate masked and unmasked codebooks (ablation)PUT · FFHQ, mask ratio 20-40%FID: 12.784 -> 12.112Mixed
remove the decoder reference branch (ablation)PUT · FFHQ, mask ratio 20-40%FID: 12.784 -> 15.014Hurt
use quantized vectors rather than encoder features as transformer inputs (ablation)PUT · FFHQ, mask ratio 20-40%FID: 12.784 -> 22.824Hurt
remove random quantization of transformer input features during training (ablation)PUT · FFHQ, mask ratio 20-40%FID: 12.784 -> 43.621Hurt

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