tkuri / papers

Summarizing the papers I have read (Japanese)
41 stars 1 forks source link

Accuracy Booster: Performance Boosting using Feature Map Re-calibration #332

Open tkuri opened 3 years ago

tkuri commented 3 years ago

論文概要

CNNの特徴マップのうち重要なチャネルを強調させ性能を向上させ、かつ従来手法(SENet)よりもシンプルな構造を提案している。キモは各チャンネルの代表的なデータポイントをまず見つけ利用しているところ。ResNet50をBoostしてReset152と同等の性能を発揮できるとのこと。

bib_20210124 00 https://openaccess.thecvf.com/content_WACV_2020/html/Singh_Accuracy_Booster_Performance_Boosting_using_Feature_Map_Re-calibration_WACV_2020_paper.html

Code

未確認。