TRANFORMERS BASED MULTI-LABEL IMAGE CLASSIFICATION AND NAMING

Computer Engineering

In recent years, the use of Machine Learning (ML) which is a type of Artificial Intelligence (AI) has increased rapidly. ML allows software applications to become more accurate in predicting the output without explicitly programmed to do so. Classification is a type of ML which categorizes given data into different classes and is used in speech recognition, face detection etc. Here image classification is considered. Multi-label classification is method of predicting set of labels according to attributes, objects present in the given image. Transformers are classification models which was using in Natural Language Processing (NLP). Now it is using in image classification. The transformer is given the data after feature extraction. Then it will be capable of predicting the labels in the image. Here in this paper, we are considering Vision Transformer (ViT). It is a model for multi-label image classification which can exploit the complex dependencies between visual features and labels. Here the concept of Multilayer Perceptron (MLP) is used to extract the features from the images. In this approach the vision-transformer is trained to predict the masked labels from the input given to it. Here the dataset used for training is the CIFAR-10. The core of this method is a label mask training objective. This model can represent label state explicitly during training. Thus, this model can be effectively applicable in medical image recognitions, wild animal recognitions and so on. This model can work at an efficiency of about 84% in its prediction accuracy.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.