AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Exploring Relational Context for Multi-Task Dense Prediction

David Brüggemann, Menelaos Kanakis, Anton Obukhov, Stamatios Georgoulis, Luc Van Gool

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 93 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 automated per-pair context type selection (ATRC) over the best single context typerelational context · NYUD-v2 (HRNet18)Delta_m: 1.33 -> 1.56Improved
use T-label context instead of local contextrelational context · NYUD-v2 (HRNet18)Delta_m: -1.34 -> 1.33Improved
add relational context distillation (ATRC) to multi-task baselinemulti-task network · NYUD-v2 (HRNet18)Delta_m: -1.89 -> 1.56Improved
add FPM to ATRCATRC · NYUD-v2 (HRNet18)Delta_m: 1.56 -> 3.57Improved
add ATRC with FPM vs MTI-NetATRC · NYUD-v2 (HRNet48)Delta_m: 0.15 -> 0.49Improved
add ATRC with FPM vs MTI-NetATRC · PASCAL-Context (HRNet18)Delta_m: -2.12 -> -1.98Improved
add ATRC to multi-task baselinemulti-task network · PASCAL-Context (HRNet18)Delta_m: -6.77 -> -4.45Improved
add ATRC to ASPP networkASPP · PASCAL-Context (ResNet-50)Delta_m: 1.77 -> 2.13Improved
add ATRC to branched BMTAS backboneBMTAS · PASCAL-Context (ResNet-50)Delta_m: -0.55 -> 4.53Improved
add self-attention blocks to ATRC (ablation)ATRC · NYUD-v2 (HRNet18)Delta_m: 0.87 -> 1.56Improved
use ground-truth T-label region maps instead of predicted (ablation)T-label context · NYUD-v2 (HRNet18)Delta_m: 1.33 -> 12.67Improved
use ground-truth S-label region maps instead of predicted (ablation)S-label context · NYUD-v2 (HRNet18)Delta_m: 1.07 -> 14.55Improved

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