Fine-grained zero-shot recognition with metric rescaling.

We address the problem of learning fine-grained cross-modal representations. We propose an instance-based deep metric learning approach in joint visual and textual space. On top of that, we derive a metric rescaling approach that solves a very common problem in the generalized zero-shot learning setting, i.e., classifying test images from unseen classes as one of the classes seen during training. We evaluate our approach on two fine-grained zero-shot learning datasets: CUB and FLOWERS. We find that on the generalized zero-shot classification task the proposed approach consistently outperforms the existing approaches on both datasets. We demonstrate that the proposed approach, notwithstanding its simplicity of implementation and training, is superior to all the recent state-of-the-art methods of which we are aware that use the same evaluation framework.