Skip to content

we investigate the scale alignment between pre-training and target datasets, and propose a new refined Scale Match method (termed SM+) for tiny person detection. SM+ improves the scale match from image level to instance level, and effectively promotes the similarity between pre-training dataset and target dataset.

License

Notifications You must be signed in to change notification settings

ucas-vg/Instance-level-Scale-Match

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 

Repository files navigation

Instance-level Scale Match (SM+)


Introduction

we investigate the scale alignment between pre-training and target datasets, and propose a new refined Scale Match method (termed SM+) for tiny person detection. SM+ improves the scale match from image level to instance level, and effectively promotes the similarity between pre-training dataset and target dataset.

Illustration of the difference between SM and SM+

image

While SM only considers the whole image, our SM+ focuses on every instance. The instance-level approach achieves scale match in a finer level. Our SM+ mainly consists of four steps: (1) Separation, (2) Instance processing, (3) Background processing, and (4) Combination.

Visualization

image

Background based on inpainting (top) vs. Background based on new sampling (bottom). The inpainting method might not repair some artifacts, but changing the background does not cause this problem.

Coming Soon!!

About

we investigate the scale alignment between pre-training and target datasets, and propose a new refined Scale Match method (termed SM+) for tiny person detection. SM+ improves the scale match from image level to instance level, and effectively promotes the similarity between pre-training dataset and target dataset.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published