/TedTorch

Deep Learning blocks and architectures built using abstract classes for ease of configuration

Primary LanguagePython

TedTorch

Pytorch Deep Learning blocks and architectures built using abstract classes for ease of configuration

Abstract classes:

  • AbstractBaseBlock
    • AbstractBaseArchitecture
  • AbstractBaseClassifier
  • AbstractBaseSequential

Configurable blocks:

  • Convolution blocks: nn.Conv2d or CoordConv
  • Activation function: Mish, Swish, ESwish, Mila
  • Norm: BatchNorm2d, InstanceNorm2d

Supported Architectures:

  • Densenet
  • Deep Aggregation Layer

Ultilities:

  • Flatten layer
  • Tedquential: extension of nn.Sequential with configurable blocks and skip connections: dense and residual