Towards few-shot object detection in the open world

PhD Qualifying Examination


Title: "Towards few-shot object detection in the open world"

by

Mr. Qi FAN


Abstract:

Conventional object detection methods for object detection typically require a 
substantial amount of training data and preparing such high-quality training 
data is very labor-intensive. This has motivated the recent development of 
few-shot object detection. Central to object detection given only a few shots 
is how to localize an unseen object in a cluttered background, which in 
hindsight is a general problem of object localization from a few annotated 
examples in novel categories. Potential bounding boxes can easily miss unseen 
objects, or else many false detections in the background can be produced. This 
task is very challenging given large variance of illumination, shape, texture, 
etc, in real-world objects.

In this survey, we give a comprehensive review of recent few-shot object 
detection methods. We first introduce a general few-shot object detection model 
that can be applied to detect novel objects without re-training and fine-tuning 
by exploiting matching relationship between object pairs in a weight-shared 
network at multiple network stages. Then we propose a technique to discover 
potential novel objects to calibrate multiple detection stages for better 
generalization on novel classes. Next, based on our analysis on the training 
data in few-shot object detection, we build a large-scale few-shot object 
detection with numerous categories and introduce extra classification dataset 
to facilitate the model training. Finally, we extend our methods to other 
related tasks and give several future research directions.


Date:			Thursday, 7 October 2021

Time:                  	4:00pm - 6:00pm

Zoom meeting:		https://hkust.zoom.us/j/8626476073

Committee Members:	Prof. Chi-Keung Tang (Supervisor)
 			Prof. Yu-Wing Tai (Supervisor)
 			Dr. Dan Xu (Chairperson)
 			Dr. Qifeng Chen


**** ALL are Welcome ****