Subset Feature Learning for Fine-Grained Category Classification

TL;DR

This paper introduces a subset feature learning approach for fine-grained classification that clusters similar classes and learns specialized features, significantly improving accuracy without relying on bounding boxes at test time.

Contribution

It proposes a novel subset clustering and deep feature learning method for fine-grained categorization, outperforming existing approaches on challenging datasets.

Findings

Achieves 77.5% accuracy on Caltech-UCSD bird dataset

Outperforms previous best of 73.2% accuracy

Demonstrates effectiveness of progressive transfer learning

Abstract

Fine-grained categorisation has been a challenging problem due to small inter-class variation, large intra-class variation and low number of training images. We propose a learning system which first clusters visually similar classes and then learns deep convolutional neural network features specific to each subset. Experiments on the popular fine-grained Caltech-UCSD bird dataset show that the proposed method outperforms recent fine-grained categorisation methods under the most difficult setting: no bounding boxes are presented at test time. It achieves a mean accuracy of 77.5%, compared to the previous best performance of 73.2%. We also show that progressive transfer learning allows us to first learn domain-generic features (for bird classification) which can then be adapted to specific set of bird classes, yielding improvements in accuracy.