Generative, Discriminative, and Ensemble Learning on Multi-Modal Perceptual Fusion Toward News Video Story Segmentation

Winston H.-M. Hsu and Shih-Fu Chang

ICME 2004 Reviews' comments :

>> Reviewer 1

Very good paper.


>> Reviewer 2

In this paper, the author describes a novel approach to news
videos segmentation. Author's segmentation method is based on
earlier work using a Maximum Entropy model.

The author proposed this new and alternative approach based on
a discriminative model, ensemble learning as well as the
original idea of proposing a method that uses combination of
ME classifiers and the associated confidence scores in each
boosting iteration.

Author also compared results of SVM-based, ME-based with pure
boosting approach, indicated in his experiment section showing
promising result.

>> Reviewer 3

It is not a surprise that BST-BIN-35 did not perform
well as each classifier is based on single binary
feature i.e. they are too weak to be boosted, let
alone the cost of 35 training iterations.

Some description on the video story types (a)-(i)
and the reason of the performance variation would
be helpful.

Fig. 2 is hard to read as there are too many curves.