In this repository, I implement computer vision techniques and ideas that are initially new to me in notebooks.
Then, if the notebook contains blocks of code that can be re-used for different vision tasks, I put them in files
under the src
directory.
In this iterative fashion, I'm building a mini vision library built off of pure PyTorch.
This repository goes from Lenet-5, to diffusion.
mnist_lenet.ipynb
: training LeNet-5 on MNIST.autoencoder.ipynb
: autoencoders.dcgan.ipynb
: Deep convolutional Generative Adversarial Networks.resnets.ipynb
: residual networks.fcn_Segmentation.ipynb
: fully convolutional networks for semantic segmentation.ddpm_mnist.ipynb
: implementing diffusion models architecture on MNIST.ddpm_cifar.ipynb
: applying the DDPM architecture fromsrc
to CIFAR-10.mixture_of_experts.ipynb
: implementing a sparsely-gated mixture of experts with ResNets.