A Cluster Method for Labelling Large Scale Vehicle Model Dataset based on Deep Learning

摘要

Vehicle make and model recognition is an important task for vehicle analysis, which can be used for automatic toll collection and public security. Vehicle make and model recognition (MMR) is a challenging task due to the close appearance between vehicle models. In this sense, MMR is a fine-grained classification task. To train a model for this finegrained task, a large scale dataset is required. Labelling a large dataset with many classifies is difficult and tedious. In this paper, we proposed a cluster method incorporating with deep learning to assist faster labelling of a large scale vehicle make and model dataset. The proposed cluster method can automatically divide the images into groups, within each group, there is one or few classes. In addition, the cluster is performed in an incremental manner to refine the deep models. Experimental results show that our cluster method speedup the label task significantly compared with manually label of each image.