I extract the multiview features of CUB dataset. It contains two views, one of which is from image features extracted by GoogleNet(2048-dimension) and the other is from text features using doc2vec(300-dimension).
The feature extraction method refer to the paper StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks by Han Zhang*, Tao Xu*, Hongsheng Li, Shaoting Zhang, Xiaogang Wang, Xiaolei Huang, Dimitris Metaxas.
You can download the processed data(has extract two views feature and save the feature in .mat file) or download the raw data from CUB's official website then run the CUB-feature_extraction.py.