Learning Semantic Context towards Pixel-Wise Recognition

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Learning Semantic Context towards Pixel-Wise Recognition"

By

Mr. Mingmin ZHEN


Abstract:

Pixel-wise recognition tasks, such as semantic segmentation and salient
object detection, aim to classify each pixel into predefined caterories.
Traditional methods suffer from the poor disriminative ability of
hand-crafted features. In this work, we delve into the CNN based methods
to advance semantic segmentation and salient object dection.

For semantic segmentation, we firstly introduce a  fully dense  neural
network with an encoder-decoder structure that we abbreviate as FDNet, in
which  feature maps of  all the previous blocks are adaptively aggregated
to feedforward as  input. On the one hand, it reconstructs  the spatial
boundaries accurately. On the other hand, it learns more efficiently  with
the more efficient  gradient backpropagation. We then present an  joint
multi-task learning framework for semantic segmentation and semantic
boundary detection. The critical component in the framework is the
iterative pyramid context module (PCM), which couples two tasks and stores
the shared latent semantics to interact between the two tasks.  A novel
loss function originated from the dual constraint  is designed to improve
further the performance for semantic segmentation, which ensures the
consistency between semantic mask boundary and boundary groundtruth.

For salient object detection, we illustrate an end-to-end differentiable
morphalogical actve contour model, which iteratively helps to improve the
boundary accuracy of salient object.  As it is hard to determine the
salient object from just one view, we also propose to obtain accurate
salient object mask from multiple views, whichs adopts feature clustering
method to correlate features from multiple views and enforces the
consistency of salient object from different views.


Date:                   Friday, 6 March 2020

Time:                   3:00pm - 5:00pm

Zoom Meeing:            https://hkust.zoom.com.cn/j/786199060

Chairman:               Prof. Ye QI (PPOL)

Committee Members:      Prof. Long QUAN (Supervisor)
                        Prof. Qifeng CHEN
                        Prof. Chiew Lan TAI
                        Prof. Kai TANG (MAE)
                        Prof. Pong Chi YUEN (HKBU)


**** ALL are Welcome ****