Getting better images from DTD using PGAN

I am trying to generate some images using the pre-trained DTD on PGAN but the results are not looking very good. Here is the pseudo-code that I am using as well as the results. Any thought on why the resulting images are not looking good? Thanks in advance

import torch
use_gpu = True if torch.cuda.is_available() else False
model = torch.hub.load('facebookresearch/pytorch_GAN_zoo:hub',
                       'PGAN', model_name='DTD',
                       pretrained=True, useGPU=use_gpu)
num_images = 4
noise, _ = model.buildNoiseData(num_images)
with torch.no_grad():
    generated_images = model.test(noise)

# let's plot these images using torchvision and matplotlib
import matplotlib.pyplot as plt
import torchvision
grid = torchvision.utils.make_grid(generated_images.clamp(min=-1, max=1), scale_each=True, normalize=True)
plt.imshow(grid.permute(1, 2, 0).cpu().numpy())
plt.show()

[Molugan]

In other words, the best way to know if a class gives good enough results is just to try it out 😃

[Molugan]

Other classes works as well, for example striped will give that kind of results:

or cracked

DTD is a really small dataset and thus the results can be very unequal. It’s more like a demo GAN showing that you can get fancy results with few data (the swirly class for example is kind of funny).