ECCV 2022 Self-Support Few-Shot Semantic Segmentation

⭐转载自ECCV 2022 | SSP: 自支持匹配的小样本任务新思想
这是paper作者自己写的,非常清楚,推荐阅读~(膜拜大佬)

Summary

1.多了self-support过程。即用M1-hat再次与query-feature提取prototype(包括前景和后景),然后与support的prototype结合得到新的support-prototype,与query-feature计算相似度得到最终的M2-hat。
2.ASBP和SSFP中提取prototype的原理,两者是不同的
3.没有额外参数

Annotation

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第1张图片

FSS Task

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第2张图片

FSS Problem

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第3张图片

Other Solutions

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第4张图片

Our Idea

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第5张图片

Our Motivation

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第6张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第7张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第8张图片

Method

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第9张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第10张图片

Another Problem ans Solution

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第11张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第12张图片

Experiments

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第13张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第14张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第15张图片
ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第16张图片

Advantages

ECCV 2022 Self-Support Few-Shot Semantic Segmentation_第17张图片

你可能感兴趣的:(FSS,Paper,Notes,深度学习)