Unconditional icon generation using DDPM Diffusion

Data set used for training: Kaggle - Icons-50

Reference code for training: Kaggle - Generating Fake Icons using DDPM Diffusion

Image samples

image/png

Sample code for use in Google Colab

Install the necessary packages

!python3 -m pip install diffusers==0.21.* accelerate

Generate images using DDPM Pipeline

from matplotlib import pyplot as plt
from diffusers import DDPMPipeline

ddpm = DDPMPipeline.from_pretrained("mayur7garg/ddpm-fake-icons")
ddpm.to("cuda")

inference_image = ddpm(
    batch_size = 36,
    num_inference_steps = 1000
).images

plt.figure(figsize = (18, 18), dpi = 120)

for i, image in enumerate(inference_image):
    plt.subplot(6, 6, i + 1)
    plt.imshow(image)
    plt.axis(False)
plt.tight_layout()
plt.show()