PVT - Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions

[junxnone]

Reference

• 2021-02 [PVT] Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions[Paper] [Code]

Brief

• Pyramid Vision Transformer
• Backbone

Motivation

• 背景/存在的问题
• 做了什么尝试/分别有什么效果
• 应用领域

Arch

PVT - Models(Tiny/Small/Medium/Large)

Evaluation

• Classification
• Object Detection
• Semantic Segmentation
• Instance Segmentation

Tricks