AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Semantic-Guided Multi-Attention Localization for Zero-Shot Learning

Yizhe Zhu, Jianwen Xie, Zhiqiang Tang, Xi Peng, Ahmed Elgammal

NeurIPS 2019 · Detection and segmentation › Open-set and few-shot detection · approach: Zero-shot segmentation and detection · 79 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 localized part patches to original image (ablation)SGMA zero-shot model · ZSL PS averagemean class accuracy (%): 59.8 -> 65.2Improved
use parts only without original image (ablation)SGMA zero-shot model · ZSL PS averagemean class accuracy (%): 59.8 -> 52.1Hurt
add random crops instead of localized parts (ablation)SGMA zero-shot model · ZSL PS averagemean class accuracy (%): 59.8 -> 57.5Hurt
add class-center triplet loss and combine both inference branches (ablation)SGMA zero-shot model · ZSL PS averagemean class accuracy (%): 59.8 -> 63.7Improved
class-center triplet inference instead of embedding softmax inference (ablation)SGMA zero-shot model · ZSL PS averagemean class accuracy (%): 60.2 -> 62.6Improved
add class-center triplet loss, softmax-branch inference only (ablation)SGMA zero-shot model · ZSL PS averagemean class accuracy (%): 59.8 -> 60.2Improved
add multi-attention loss (compactness and diversity) (ablation)SGMA zero-shot model · CUB part detectionaverage precision (%): 47.6 -> 61.5Improved
share CNN weights between image and part patches (ablation)SGMA zero-shot model · CUB PSmean class accuracy (%): -3.6Hurt
share CNN weights between image and part patches (ablation)SGMA zero-shot model · AWA PSmean class accuracy (%): -2.7Hurt
increase number of attention parts beyond 2 (ablation)SGMA zero-shot model · CUB, AWA, FLOmean class accuracy (%)No clear effect
SGMA instead of LDF in generalized ZSLSGMA zero-shot model · CUB GZSLharmonic mean H (%): 39.9 -> 48.5Improved
SGMA vs LDF on seen-class accuracySGMA zero-shot model · CUB GZSLA_S (%): 81.6 -> 71.3Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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