Texture Extraction Methods Based Ensembling Framework for Improved Classification
Authors:
Vijay Pandey,
Trapti Kalra,
Mayank Gubba,
Mohammed Faisal
Abstract:
Texture-based classification solutions have proven their significance in many domains, from industrial inspections to health-related applications. New methods have been developed based on texture feature learning and CNN-based architectures to address computer vision use cases for images with rich texture-based features. In recent years, architectures solving texture-based classification problems…
▽ More
Texture-based classification solutions have proven their significance in many domains, from industrial inspections to health-related applications. New methods have been developed based on texture feature learning and CNN-based architectures to address computer vision use cases for images with rich texture-based features. In recent years, architectures solving texture-based classification problems and demonstrating state-of-the-art results have emerged. Yet, one limitation of these approaches is that they cannot claim to be suitable for all types of image texture patterns. Each technique has an advantage for a specific texture type only. To address this shortcoming, we propose a framework that combines more than one texture-based techniques together, uniquely, with a CNN backbone to extract the most relevant texture features. This enables the model to be trained in a self-selective manner and produce improved results over current published benchmarks -- with almost same number of model parameters. Our proposed framework works well on most texture types simultaneously and allows flexibility for additional texture-based methods to be accommodated to achieve better results than existing architectures. In this work, firstly, we present an analysis on the relative importance of existing techniques when used alone and in combination with other TE methods on benchmark datasets. Secondly, we show that Global Average Pooling which represents the spatial information -- is of less significance in comparison to the TE method(s) applied in the network while training for texture-based classification tasks. Finally, we present state-of-the-art results for several texture-based benchmark datasets by combining three existing texture-based techniques using our proposed framework.
△ Less
Submitted 20 October, 2022; v1 submitted 8 June, 2022;
originally announced June 2022.
Maximum weighted independent sets with a budget
Authors:
Tushar Kalra,
Rogers Mathew,
Sudebkumar Prasant Pal,
Vijay Pandey
Abstract:
Given a graph $G$, a non-negative integer $k$, and a weight function that maps each vertex in $G$ to a positive real number, the \emph{Maximum Weighted Budgeted Independent Set (MWBIS) problem} is about finding a maximum weighted independent set in $G$ of cardinality at most $k$. A special case of MWBIS, when the weight assigned to each vertex is equal to its degree in $G$, is called the \emph{Max…
▽ More
Given a graph $G$, a non-negative integer $k$, and a weight function that maps each vertex in $G$ to a positive real number, the \emph{Maximum Weighted Budgeted Independent Set (MWBIS) problem} is about finding a maximum weighted independent set in $G$ of cardinality at most $k$. A special case of MWBIS, when the weight assigned to each vertex is equal to its degree in $G$, is called the \emph{Maximum Independent Vertex Coverage (MIVC)} problem. In other words, the MIVC problem is about finding an independent set of cardinality at most $k$ with maximum coverage.
Since it is a generalization of the well-known Maximum Weighted Independent Set (MWIS) problem, MWBIS too does not have any constant factor polynomial time approximation algorithm assuming $P \neq NP$. In this paper, we study MWBIS in the context of bipartite graphs. We show that, unlike MWIS, the MIVC (and thereby the MWBIS) problem in bipartite graphs is NP-hard. Then, we show that the MWBIS problem admits a $\frac{1}{2}$-factor approximation algorithm in the class of bipartite graphs, which matches the integrality gap of a natural LP relaxation.
△ Less
Submitted 19 July, 2015; v1 submitted 25 June, 2015;
originally announced June 2015.
