These experiments perform linear interpolation in the text-embedding space. In a similar way to the previous experiment, we do this by sampling two text-embedding vectors, which are represented as points in a multidimensional space. We interpolate between them by linearly going from one point to another in N steps.
The following code block is used for inference with interpolation in the text-embedding space:
def infer(data_filepath='data/flowers.hdf5', z_dim=128, out_dir='gan', n_steps=10): # we load the saved model G = load_model(out_dir) # get text embeddings and text from the validation set val_data = get_data(data_filepath, 'train') val_data = next(iterate_minibatches(val_data, 2)) # sample ...
