The aim of this repository is to get to know how images are being processed. It covers several topics, including:
- generating artificial images (using GAN and DDPM),
- image classification using visual transformer.
By employing a dataset consisting of pumpkin cakes, we train both a generator and a discriminator model. The primary objective of the generator is to generate high-quality pumpkin cakes, while the discriminator aims to distinguish between real and fake ones. In order to expedite the training process, the images are downscaled to a resolution of 32x32 pixels.
Results (upscaled)
GAN model structure
The objective of this laboratory is to educate the PyTorch model on the task of denoising images. For this purpose, we emulate an image as a vector with a shape of (2,). Initially, the model undergoes training using the bicycle.txt dataset. Subsequently, we employ the trained model to generate a denoised bicycle image by removing the noise introduced through a normal distribution.
Bike generated from noise
Image classification model using Vision Transformers.
jupyter nbconvert --to webpdf --allow-chromium-download lab.ipynb