keras implementation of paper: Feature Generating Networks for Zero-Shot Learning
Download: data This dataset provides a platform to benchmark transfer-learning algorithms, in particular, attribute base classification and zero-shot learning. It can act as a drop-in replacement to the original Animals with Attributes (AwA) dataset, as it has the same class structure and almost the same characteristics. In addition, on this website, you will find CUB, SUN, FLO.
In order to run our experiments, and facilitate visualization of the dataset, we introduce the GZSL datasets in H5file format. You can download the datasets CUB, SUN, FLO, and AWA1 in the h5file format, which can be downloaded dataset.zip (~3GB in zip file).