#191
summarized by : Anonymous
MPViT: Multi-Path Vision Transformer for Dense Prediction
Youngwan Lee; Jonghee Kim; Jeffrey Willette; Sung Ju Hwang

どんな論文か？

improving multi-scale feature representation for dense prediction tasks by multi-scale patch embedding and multi-path feature aggregation
placeholder

新規性

multi-scale embedding with a multipath structure for simultaneously representing fine and coarse features for dense prediction tasks; global to local feature interaction module

結果

SOTA results on ImageNet-1K classification, COCO detection, ADE20K segmentation compared with other backbones.

その他（なぜ通ったか？等）

このページで利用されている画像は論文から引用しています．