AI Paper Atlas

Research map / Detection and segmentation / CNN features

Distraction-Aware Feature Learning for Human Attribute Recognition via Coarse-to-Fine Attention Mechanism

Mingda Wu, Di Huang, Yuanfang Guo, Yunhong Wang

AAAI 2020 · Detection and segmentation › CNN features · approach: Local features and context in CNNs · 30 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 Da-HAR's coarse-to-fine attention and multi-level features instead of DIAAHuman attribute recognition · WIDER-AttributemAP: 86.4 -> 87.3Improved
Use Da-HAR with original BCE-lossDa-HAR · RAPPrec: 84.54Improved
Use weighted BCE-loss to emphasize minority attributesDa-HAR · RAPRec: 84.13Improved
Use weighted BCE-lossDa-HAR · RAPPrec: 84.54 -> 66.50Hurt
Use mixed loss to trade off Precision and RecallDa-HAR · RAPF1: 80.72Improved
Add multi-level feature fusion (ablation)Da-HAR · WIDER-AttributemAP: 85.3 / 86.2 -> 85.7%/86.5%Improved
Add the self-mask block for coarse distraction awareness (ablation)Da-HAR · WIDER-AttributemAP: 85.3 / 86.2 -> 85.8%/86.6%Improved
Add the masked attention block for fine distraction-aware learning (ablation)Da-HAR · WIDER-AttributemAP: 85.3 / 86.2 -> 86.0%/86.8%Improved
Combine multi-level feature fusion, self-mask, and masked attention (ablation)Da-HAR · WIDER-AttributemAP: 85.3 / 86.2 -> 86.2%/87.1%Improved
Ignore 0-label attributes during training rather than treating them as negatives (ablation)Da-HAR · WIDER-AttributemAP: 86.2%/87.1% -> 86.5%/87.3%Improved
Add validation data to training (ablation)Da-HAR · WIDER-AttributemAP: 86.5%/87.3% -> 87.2%/88.0%Improved
Add the proposed blocks to the baseline networkDa-HAR · WIDER-Attributeinference time: 22.28ms -> 24.04msHurt

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