Title: Few-Shot Learning for Semantic Segmentation

Abstract: 

Most learning architectures for segmentation task require a significant amount of data and annotations, especially in the task of segmentation, where each pixel is assigned to a class. Few-shot segmentation aims to replace large amount of training data with only a few densely annotated samples. In this talk, I’ll present a two-branch network, FuseNet, that can few-shot segment an input image given one or multiple images of the target domain. We also show the quantitative effects of number of training samples on Intersection over Union(IoU). Our network achieves the state-of-the-art result on PASCAL VOC 2012 for both one-shot and five-shot semantic segmentation.

Committee:

Martial Hebert (advisor)

Jean Oh (co-advisor)

Deva Ramanan

Allison Del Giorno