PyTorch Land

This is a repository of select implementations of deep learning models using pytorch.

There is also my own mini-framework for model training (which I call simply NetworkTrainer), that ended up with something very similar to ignite, created in order to reduce common boilerplating. More flexible and intuitive than ignite, in my opinion :).

More models to be added, and improvements on NetworkTrainer is under way.

Installation

pip3 install pytorch-land

See pypi page for package details.

Implemented Models

CNN architectures

• ResNet (2015) [paper] [code]
• Inception v.2 (2015) [paper] [code]
• MobileNet (2017) [paper] [code]

GANs

• GAN (2014) [paper] [code]
• DCGAN (2015) [paper] [code]
• InfoGAN (2016) [paper] [code]
• f-GAN (2016) [paper] [code]
• UnrolledGAN (2016) [paper] [code] [train-examples]
• ACGAN (2016) [paper] [code] [train-examples]
• BEGAN (2017) [paper] [code & examples]
• CycleGAN (2017) [paper] [code & examples]

Autoencoders

• Stacked Denoising Autoencoders [paper] [code]
• Stacked Convolutional Denoising Autoencoders (2017) [paper] [code]

Requirements

Required packages are specified in requirements.txt file. The packages can be installed using the following command:

pip3 install -r requirements.txt

Notably, the codes are compatible with pytorch 0.4 - working on with pytorch 1.1 compatibility.

NetworkTrainer

Datasets