torchkit

PyTorch Utility Package to setup training and testing pipeline for Computer Vision Tasks

File Structure

Package has 5 sub-packages

1. data

Consists of Dataset, Dataloader functions and classes

2. models

Has two different network files, based on CIFAR-10 and MNIST

3. run

Consists of Train and Testing part of NeuralNet

4. torchsummary

Mainly modelsummary with Receptive Field calculated layer-wise

5. utils

Consists of DataUtils and ModelUtils, which has helper functions mainly to plot and visualize data in former, & latter has model related functions.

Features

Convolutions

* Depthwise
* Dilated 

Normalization

* BatchNorm
* GroupNorm
* LayerNorm

Model Summary

* with layer-wise Receptive Field

Model utilities

Loss functions

    * Cross Entropy Loss
    * NLLoss

Evaluation Metrics

    * Accuracy

Optimizers

    * Stochastic Gradient Descent

LR Schedulers

    * Step LR
    * Reduce LR on Plateau
    * One Cycle Policy

Datasets

* MNIST
* CIFAR10