Subset Feature Learning for Fine-Grained Category Classification

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 pro-pose a learning system which first clusters visually similar classes and then learns deep convolutional neural network features specific to each subset. Experiments on the popu-lar fine-grained Caltech-UCSD bird dataset show that the proposed method outperforms recent fine-grained categori-sation methods under the most difficult setting: no bounding boxes are presented at test time. It achieves a mean accu-racy 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. 1