torch-tensornet 0.5

A high-level deep learning library build on top of PyTorch.

TensorNet

TensorNet is a high-level deep learning library built on top of PyTorch.

NOTE: This documentation applies to the MASTER version of TensorNet only.

Installation

You can use pip to install tensornet

pip install torch-tensornet

If you want to get the latest version of the code before it is released on PyPI you can install the library from GitHub

pip install git+https://github.com/shan18/TensorNet.git#egg=torch-tensornet

Features

TensorNet currently supports the following features

• Model architectures
  • ResNet18
  • A custom model called BasicNet
• Model utilities
  • Loss functions
    • Cross Entropy Loss
    • Binary Cross Entropy Loss
    • Mean Square Error Loss
    • SSIM and MS-SSIM Loss
    • Dice Loss
  • Evaluation Metrics
    • Accuracy
    • RMSE
    • MAE
    • ABS_REL
  • Optimizers
    • Stochastic Gradient Descent
  • Regularizers
    • L1 regularization
    • L2 regularization
  • LR Schedulers
    • Step LR
    • Reduce LR on Plateau
    • One Cycle Policy
  • LR Range Test
  • Model Checkpointing
  • Tensorboard
• Model training and validation
• Datasets (data is is returned via data loaders)
  • MNIST
  • CIFAR10
  • TinyImageNet
  • MODEST Museum Dataset
• Data Augmentation
  • Resize
  • Padding
  • Random Crop
  • Horizontal Flip
  • Vertical Flip
  • Gaussian Blur
  • Random Rotation
  • CutOut
• GradCAM and GradCAM++ (Gradient-weighted Class Activation Map)
• Result Analysis Tools
  • Plotting changes in validation accuracy and loss during model training
  • Displaying correct and incorrect predictions of a trained model

How to Use

For examples on how to use TensorNet, refer to the examples directory.

Dependencies

TensorNet has the following third-party dependencies

• torch
• torchvision
• torchsummary
• tqdm
• matplotlib
• albumentations
• opencv-python

Documentation

Documentation making for the library is currently in progress. So until a documentation is available please refer to the following table for various functionalities and their corresponding module names.

Functionality	Module Name
Training, Validation and LR scheduling	engine
Dataset downloading and preprocessing	data
GradCAM and GradCAM++	gradcam
Models, loss functions and optimizers	model
CUDA setup and result analysis	utils

For a demo on how to use these modules, refer to the notebooks present in the examples directory.

Contact/Getting Help

If you need any help or want to report a bug, raise an issue in the repo.