Ross Luo and Michael Laielli and Giscard Biamby and Adam Loeffler

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2020-233

December 19, 2020

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/EECS-2020-233.pdf

In recent years, the AI research and the humanitarian assistance and disaster response (HADR) communities have sought to collaborate together: There is a growing desire in the AI research community to transition state of the art research towards endeavors for social good. Likewise, the HADR community, comprised mainly of NGO's, governments, and not-for-profit entities, has historically been insulated from the latest technological advances and welcomes infusions of technical insights. Part I of this work describes the origins, progress, and takeaways of these collaborations.

Part II details some of my team's ongoing contributions in this area. Notably, we are the first to adapt structured Gaussian filters to the object detection task. We evaluate our CenterNet-DLA detector with spherical Gaussian filters on COCO and the xView overhead object detection dataset and achieve performance comparable to one with free-form deformable convolution filters while utilizing fewer dynamic parameters.

Advisors: Trevor Darrell

---

BibTeX citation:

@mastersthesis{Luo:EECS-2020-233,
    Author= {Luo, Ross and Laielli, Michael and Biamby, Giscard and Loeffler, Adam},
    Editor= {Darrell, Trevor and Canny, John F.},
    Title= {AI for HADR: Progress and Opportunities},
    School= {EECS Department, University of California, Berkeley},
    Year= {2020},
    Month= {Dec},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/EECS-2020-233.html},
    Number= {UCB/EECS-2020-233},
    Abstract= {In recent years, the AI research and the  humanitarian assistance and disaster response (HADR) communities have sought to collaborate together: There is a growing desire in the AI research community to transition state of the art research towards endeavors for social good. Likewise, the HADR community, comprised mainly of NGO's, governments, and not-for-profit entities, has historically been insulated from the latest technological advances and welcomes infusions of technical insights. Part I of this work describes the origins, progress, and takeaways of these collaborations. 

Part II details some of my team's ongoing contributions in this area. Notably, we are the first to adapt structured Gaussian filters to the object detection task. We evaluate our CenterNet-DLA detector with spherical Gaussian filters on COCO and the xView overhead object detection dataset and achieve performance comparable to one with free-form deformable convolution filters while utilizing fewer dynamic parameters.},
}

EndNote citation:

%0 Thesis
%A Luo, Ross 
%A Laielli, Michael 
%A Biamby, Giscard 
%A Loeffler, Adam 
%E Darrell, Trevor 
%E Canny, John F. 
%T AI for HADR: Progress and Opportunities
%I EECS Department, University of California, Berkeley
%D 2020
%8 December 19
%@ UCB/EECS-2020-233
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/EECS-2020-233.html
%F Luo:EECS-2020-233