CS Events

PhD Defense

Maximize Utilization of Support-Set for Few-shot segmentation

 

Download as iCal file

Wednesday, May 01, 2024, 03:00pm

 

Speaker: Seonghyeon Moon

Location: CBIM 22 Multipurpose Room

Committee Composition:

  • Professor Mubbasir Kapadia from Rutgers University
  • Professor Vladimir Pavlovic from Rutgers University
  • Professor Mridul Aanjaneya from Rutgers University
  • Professor Xia Haifeng from Southeast University (external)

Abstract:

In Few-shot segmentation(FSS), the support set plays a critical role as it provides target information to segment the target object in a given image. However, previous works focused on improving network architecture to get performance improvements neglecting the importance of how to utilize target features from the support set. We observed there were performance bottlenecks because of the limited utilization of the support set. We investigated information loss and proposed new approaches to retrieve this loss to get more meaningful information from the support set. We validated the effectiveness of our approach by instantiating it into three recent and strong FSS methods. Experimental results on several publicly available FSS benchmarks show that our proposed methods consistently improve performance by visible margins and lead to faster convergence.

Zoom:

https://rutgers.zoom.us/j/6756059091?pwd=M0RoNjY4YWN1OWptNys5WlRqMFhmdz09&omn=91204206398
Join by SIP
6756059091@zoomcrc.com
Meeting ID: 675 605 9091
Password: 077449

Speaker: Seonghyeon Moon

Event Type: PhD Defense