torch-soft 0.1.1

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

torch_soft is a high level implementatio of pytorch for some of claasification implementation with ease...

features

TensorNet currently supports the following features

• Model architectures
  • ResNet18
  • A custom model called naiveresnet
• Model utilities
  • Loss functions
    • Cross entropy loss
    • nll_loss
  • Optimizers
    • Stochastic Gradient Descent
  • Regularizers
    • L1 regularization
    • L2 regularization
  • LR Schedulers
    • Step LR
    • Reduce LR on Plateau
    • One Cycle Policy
  • LR Range Test
• Model training and validation
• Datasets (data is is returned via data loaders)
  • MNIST
  • CIFAR10
  • TinyImageNet
• 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
  • Plotting images in a batch for visualization
  • Plotting gradcam outputs

How to Use

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

Dependencies

torch_soft has the following third-party dependencies

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