AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

MTFormer: Multi-task Learning via Transformer and Cross-Task Reasoning

Xiaogang Xu, Hengshuang Zhao, Vibhav Vineet, Ser-Nam Lim, Antonio Torralba

ECCV 2022 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 76 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 cross-task attention (ablation)MTFormer (no CL) · NYUD-v2delta_m %: 2.87 -> 3.56Improved
add cross-task attention (ablation)MTFormer (no CL) · PASCAL-Contextdelta_m %: 1.55 -> 1.91Improved
add cross-task contrastive learning (ablation)MTFormer · NYUD-v2delta_m %: 3.56 -> 4.12Improved
add cross-task contrastive learning (ablation)MTFormer · PASCAL-Contextdelta_m %: 1.91 -> 2.41Improved
add feature mapping before MLP fusion (ablation)MTFormer (no CL) · NYUD-v2delta_m %: 3.49 -> 3.56Improved
add feature mapping before MLP fusion (ablation)MTFormer (no CL) · PASCAL-Contextdelta_m %: 1.58 -> 1.91Improved
add MLP fusion instead of summation (ablation)MTFormer (no CL) · NYUD-v2delta_m %: 3.38 -> 3.56Improved
add feature balance (smaller cross-task channels) (ablation)MTFormer (no CL) · PASCAL-Contextdelta_m %: 1.70 -> 1.91Improved
use transformer instead of CNN for multi-task learningMTL · NYUD-v2delta_m %: -2.61 -> +2.87Improved
use transformer instead of CNN for multi-task learningMTL · PASCAL-Contextdelta_m %: -4.96 -> +1.55Improved
share a deep decoder with lightweight branches (MTL vs STL params)Transformer MTL · NYUD-v2Params (M): 114.26 -> 62.45Improved
add MTL with few-shot part annotationsMTFormer · PASCAL few-shot (1% part labels)Human part mIoU: 11.27 -> 51.74Improved

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