AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

FuseFormer: Fusing Fine-Grained Information in Transformers for Video Inpainting

Rui Liu, Hanming Deng, Yangyi Huang, Xiaoyu Shi, Lewei Lu, Wenxiu Sun, Xiaogang Wang, Jifeng Dai, Hongsheng Li

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 173 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 soft split and soft composition (overlapping patches, patch 7) (ablation)ViB-T · Table 1 ablationPSNR: 30.50 -> 31.02Improved
add soft composition only (overlap in composition) (ablation)ViB-S · Table 1 ablationPSNR: 30.50 -> 30.74Improved
add soft split only (ablation)ViB-S · Table 1 ablationPSNR: 30.50 -> 30.99Improved
use both soft split and soft composition rather than soft split only (ablation)ViB-S · Table 1 ablationPSNR: 30.99 -> 31.02Improved
increase ViB-S patch size from 5 to 7 (ablation)ViB-S · Table 1 ablationPSNR: 30.91 -> 31.02Improved
increase ViB-T patch size from 3 to 7 without overlap (ablation)ViB-T · Table 1 ablationPSNR: 30.68 -> 30.50Hurt
replace standard Transformer block with FuseFormer block (F3N sub-token fusion) (ablation)ViB-S · Table 1 ablationPSNR: 31.02 -> 31.72Improved
add normalization in F3N (ablation)ViF · Table 1 ablationPSNR: 31.72 -> 31.87Improved
add normalization in F3N (ablation)ViF · Table 1 ablationSSIM: 0.9654 -> 0.9662Improved
FuseFormer vs STTN on video completion (ablation)ViF · YouTube-VOSPSNR: 32.34 -> 33.16Improved
FuseFormer vs STTN on video completion (ablation)ViF · DAVISPSNR: 30.67 -> 32.54Improved
replace Transformer block with FuseFormer block, VFID (ablation)ViB-S · YouTube-VOSVFID: 0.056 -> 0.051Improved

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