Multi-label and Multi-class Classification on a Custom Dataset using Convolution Neural Networks
Sridevi Bonthu, Kompella Bhargava Kiran, Mamatha Deenakonda, V. V. R. Maheswara Rao, S. Jagadeesh
Abstract
A multi-label image classification is a challenging task as it has to map an input image to a vector of outputs. This work presents a single and efficient model to perform multi-label and multi-class classification using Convolutional Neural Networks (CNN). A custom dataset is collected and annotated with eight labels to test the neural network. The classes for the labels in the dataset are a blend of balanced and unbalanced data. The proposed network is a two-stage network. The first stage is used for general training, whereas the second stage is used for specific training. The results show that the usage of a single model across multiple labels results in varying performance. The number of classes and the balance in the class distribution have a direct impact on the performance of the model. The code is available at https://github.com/sridevibonthu/FML.