KalelPark's LAB

[ Computer Vision ] Sparsely Embedded Convolution이란? 본문

Data Science/Common

[ Computer Vision ] Sparsely Embedded Convolution이란?

kalelpark 2023. 3. 25. 15:00

기존 3D point cloud와 관련된 방법론들은, grid-based method 방식이 대다수였습니다.
grid-based 방식을 크게 발전시킨 방법으로는, SP Conv입니다.

 

Sparse Convd 는 Computing cost를 매우 줄일 수 있는 방식으로, 실시간 detection에서 매우 효과적입니다.

기존 Grid-based Method는 point cloud에서는 대부분의 정보가 비어있기 때문에, 적합하지 않습니다. 

Sparse Conv는 매우 sparse한 data인 point cloud의 문제를 해결하는데 매우 효과적인 방식입니다.

Rule book 즉, 정보가 존재하는 Rule에 따라, 해당부분만을 계산하여, 정보를 처리하는 연산입니다.

이러한 원리를 적용한다면 속도가 상당히 빠르다는 것을 알 수 있습니다. 또한, 속도와 성능이 상당히 높다는 것을 알 수 있습니다.

Reference

https://arxiv.org/abs/1711.10275

 

3D Semantic Segmentation with Submanifold Sparse Convolutional Networks

Convolutional networks are the de-facto standard for analyzing spatio-temporal data such as images, videos, and 3D shapes. Whilst some of this data is naturally dense (e.g., photos), many other data sources are inherently sparse. Examples include 3D point

arxiv.org

https://www.mdpi.com/1424-8220/18/10/3337

 

SECOND: Sparsely Embedded Convolutional Detection

LiDAR-based or RGB-D-based object detection is used in numerous applications, ranging from autonomous driving to robot vision. Voxel-based 3D convolutional networks have been used for some time to enhance the retention of information when processing point

www.mdpi.com

https://towardsdatascience.com/how-does-sparse-convolution-work-3257a0a8fd1

 

How does sparse convolution work?

Sparse convolution plays an essential role for LiDAR signal processing. This article describes how the sparse convolution works.

towardsdatascience.com

https://dh87.tistory.com/3

 

SECOND: Sparsely Embedded Convolutional Detection 겉핥기 리뷰

※ 아이디어 위주로 논문을 겉핥은 글입니다. 오류가 있으면 댓글로 알려주세요 #1 SECOND 논문의 의의 Point Net의 발표이후 point cloud를 이용한 3d obect detection에 대한 논문들이 연이어 발표되었다.

dh87.tistory.com

 

Comments